All signals

Data center lobby secured EU provision to keep environmental impact data confidential - report

A new report reveals that a lobbying group representing major tech companies, including Microsoft, Amazon, Google, and Meta, successfully secured a provision in the EU to keep data center environmental impact data confidential.

EU Commission awards 180 million euro cloud contract to four European providers

The European Commission has awarded a €180 million cloud computing contract to four European providers, aiming to bolster sovereign infrastructure.

Anthropic CEO met White House chief of staff as US seeks access to Mythos model

The CEO of Anthropic met with the White House chief of staff as the US government seeks access to evaluate the company's new Mythos model.

Second meeting of Special Panel to advise President von der Leyen on child online safety

A special panel convened to advise European Commission President Ursula von der Leyen on strategies for ensuring child online safety.

Anthropic publishes Claude Opus 4.7 system card

Crypto and AI Pacs raise $250mn ahead of US midterm elections

Political action committees representing the crypto and AI industries have raised $250 million ahead of the US midterm elections.

States can—and should—regulate AI in criminal justice

A Brookings Institution article argues that U.S. states have the authority and responsibility to regulate the use of AI in the criminal justice system.

Commission proposes measures to Google on sharing search engine data with third parties under Digital Markets Act

The European Commission proposed measures under the Digital Markets Act to mandate Google's sharing of search engine data with third parties.

Jane Doe: OpenAI acknowledged abuse report but took no action

In Jane Doe v. OpenAI (TRO April 13 2026), plaintiff filed formal Nov 13 2025 Notice of Abuse; OpenAI called report "extremely serious and troubling" but took no product action before further harm.

Hot Off the Press: EFF's Updated Guide to Tech at the US-Mexico Border

The Electronic Frontier Foundation published an updated guide to help journalists and advocates identify and report on surveillance technology deployed at the U.S.-Mexico border.

Stanford AI Index 2026 reports declining public trust in AI companies

Jane Doe v. OpenAI: TRO entered in SF Superior Court after ChatGPT-fueled stalking

SF Superior Court entered a limited TRO in Doe v. OpenAI (CGC-26-635725) on April 13 2026 after plaintiff alleged ChatGPT facilitated her stalker despite abuse reports; liability theory advances.

Florida AG opens OpenAI investigation over undisclosed incidents

Comparison Shopping Is Not a (Computer) Crime

A federal district court ruled that Perplexity's AI-enabled browser violated the Computer Fraud and Abuse Act by accessing Amazon's website to help users comparison shop. The EFF has filed an amicus brief supporting Perplexity's appeal to the Ninth Circuit, arguing the decision harms competition and innovation.

Florida AG opens OpenAI consumer-protection investigation

Florida AG James Uthmeier opened a consumer-protection investigation into OpenAI on April 9 2026 focused on deceptive claims and harm to minors; joins Texas AG actions.

OpenAI Child Safety Blueprint published

OpenAI published Child Safety Blueprint on April 8 2026 detailing reporting channels, provider coordination, and safety-by-design controls for AI-enabled child exploitation.

Digital Hopes, Real Power: How the Arab Spring Fueled a Global Surveillance Boom

The Electronic Frontier Foundation published a blog series reflecting on how the 2011 Arab uprisings inadvertently fueled a global boom in state surveillance, including the rise of AI-driven biometrics and facial recognition.

Anthropic withholds Claude Mythos Preview but publishes disclosure rationale

EU Parliament Blocks Mass-Scanning of Our Chats—What's Next?

The EU Parliament voted not to extend an interim derogation from e-Privacy rules, effectively making the voluntary mass-scanning of private chats by tech companies illegal in the EU.

Anthropic discloses Claude Mythos Preview cyber capabilities pre-release

Stanford AI Index 2026: public trust in AI companies declining globally

Stanford HAI AI Index 2026 Public Opinion chapter documents declining global trust in AI companies and broad support for mandated AI disclosure across surveyed countries.

Anthropic publishes Claude Mythos Preview system card

UN opens public consultation for first Global Dialogue on AI Governance

UN opened worldwide public submission portal (deadline 30 Apr 2026) for input into the first Global Dialogue on AI Governance, to be held in 2026 in Geneva back-to-back with the ITU AI for Good Summit. Dialogue + Independent International Scientific Panel on AI established by UNGA Resolution A/RES/79/325 (adopted by consensus 26 Aug 2025). Signals formal public-expectation channel for cross-border AI governance.

Tech Nonprofits to Feds: Don’t Weaponize Procurement to Undermine AI Trust and Safety

Tech nonprofits, including the EFF and CDT, filed comments opposing a proposed GSA procurement rule that would require AI contractors to license their systems for "all lawful purposes," arguing it could enable mass surveillance.

Google and Amazon: Acknowledged Risks, and Ignored Responsibilities

The Electronic Frontier Foundation publicly criticized Google and Amazon for failing to address human rights and surveillance risks associated with their Project Nimbus AI cloud contract with the Israeli government.

EFF’s Submission to the UN OHCHR on Protection of Human Rights Defenders in the Digital Age

The Electronic Frontier Foundation (EFF) submitted a report to the UN OHCHR detailing how new digital regulations and surveillance technologies, including biometric monitoring, are being used to restrict the fundamental rights of human rights defenders globally.

Documentary "The AI Doc" elevates mainstream debate on AI well-being

CBS Mornings, NBC, AP, Oprah Podcast and others covered a new documentary featuring Tristan Harris and industry leaders on AI risks including companionship and mental-health impact, April 2026 release cycle.

Anthropic Fellows Program opens next two cohorts (May & July 2026)

Anthropic opened applications for its next Fellows cohorts beginning May and July 2026. 4-month program, $3,850/week stipend + ~$15k/mo compute funding + mentorship for independent safety researchers. 40% of first-cohort fellows joined Anthropic full-time; 80%+ produced papers.

What jobseekers need to know about automated recruitment decisions

The UK Information Commissioner's Office (ICO) published guidance for jobseekers explaining their rights and protections when facing automated recruitment decisions.

Mistral raises €722M debt to build sovereign European AI compute

Mistral AI secured €722M ($830M) in debt financing from seven-bank consortium led by Bpifrance and BNP Paribas to build a 44MW data center near Paris with 13,800 Nvidia GB300 GPUs (Q2 2026 target). Largest AI-focused debt raise by a European technology company to date; targets 200MW capacity by end-2027.

Center for Humane Technology launches "AI and What Makes Us Human" initiative

CHT announced a new program focused on what norms, legal protections, and rights are needed to preserve human well-being in the age of AI, amplifying well-being-over-engagement discourse.

NY RAISE Act adds 72-hour incident reporting requirement

NY RAISE Act chapter amendments tighten pre-deployment testing requirements

Traffic Violation! License Plate Reader Mission Creep Is Already Here

Civil liberties advocates and journalists exposed that automated license plate readers are being used for traffic enforcement, contradicting vendor claims.

👓 Who's Really Watching What Smartglasses See? | EFFector 38.6

The Electronic Frontier Foundation (EFF) published a newsletter and podcast episode raising privacy concerns about Meta Ray-Ban smartglasses and their surveillance capabilities.

EFF Sues for Answers About Medicare's AI Experiment

The Electronic Frontier Foundation filed a FOIA lawsuit against CMS seeking transparency on the AI algorithms used to evaluate Medicare prior authorization requests.

Washington HB 2225 signed: AI chatbot duty-of-care law

Governor Ferguson signed HB 2225 on March 24 2026, extending duty-of-care requirements to AI companion chatbots in Washington with private right of action; effective Jan 1 2027.

Court of Rome annuls Italy's €15M GDPR fine against OpenAI

On 18 March 2026, the Court of Rome annulled the Garante's Nov 2024 €15M fine — the only final GDPR enforcement decision against a GenAI provider. The fine had found unlawful training data processing, breach-notification failure and no age verification. Garante can appeal. Judgment no. 4153/2026, R.G. 4785/2025.

Colorado AI Act effective date pushed to June 2026 amid industry pressure to repeal

CO SB 24-205, the first US comprehensive AI anti-discrimination law, was delayed from Feb 2026 to 30 June 2026 via SB 25B-004. In March 2026, Governor Polis's AI Policy Working Group proposed a replacement bill stripping many employer compliance duties.

Minnesota introduces HF 4369 (Safeguarding Human Intelligence Act) with 90-day AI displacement notice

HF 4369 introduced 17 March 2026 requires employers to notify workers 90 days before AI-driven layoffs and fund retraining; first US state-level AI displacement notification bill. SF 4576 companion filed in Senate.

Colorado AI Act working-group proposal strips duty of care

Colorado AI Policy Work Group released March 17 2026 draft to repeal-and-replace the Colorado AI Act, eliminating its "reasonable care" duty and narrowing scope to ADMT disclosure obligations.

IDS / African Digital Rights Network report: 11 African governments spent $2B+ on Chinese-built AI surveillance

The Institute of Development Studies, with the African Digital Rights Network, published 'Smart City Surveillance in Africa: Mapping Chinese AI Surveillance Across 11 Countries' on 12 Mar 2026. Documents $2B+ spent by Algeria, Egypt, Kenya, Mauritius, Mozambique, Nigeria, Rwanda, Senegal, Uganda, Zambia and Zimbabwe on facial recognition and ANPR, with deployments used against activists, opposition figures and journalists despite no demonstrated crime-reduction effect.

Pew: 50% of Americans more concerned than excited about AI

Pew Research Center survey found 50% of US adults are more concerned than excited about AI, up from 37% in 2021; majority want stricter accountability for AI harms.

EU Parliament backs EU conclusion of the Council of Europe Framework Convention on Artificial Intelligence

The European Parliament approved the European Union's conclusion of the Council of Europe Framework Convention on Artificial Intelligence.

European Parliament approves EU accession to Council of Europe AI Convention

On 11 Mar 2026 the European Parliament gave consent (455 in favour, 101 against, 74 abstentions) to EU conclusion of the Council of Europe Framework Convention on AI and Human Rights (CETS 225), the first legally binding international AI treaty. Advances ratification; treaty enters into force after 5 states ratify.

UK Politicians Continue to Miss the Point in Latest Social Media Ban Proposal

The UK House of Commons proposed an amendment to the Children's Wellbeing and Schools Bill that would empower the Secretary of State to restrict under-18s' access to social media and addictive features.

The one question everyone should be asking after OpenAI’s deal with the Pentagon

The AI Now Institute published an article questioning the lack of safety guardrails for high-stakes decisions and surveillance following OpenAI's deal with the Pentagon.

KOSA stalled in Senate Commerce Committee under Cruz

Despite 75+ co-sponsors, KOSA has not received a markup from Sen. Cruz's committee as of Feb 2026. House advanced narrower KIDS Act (28-24, party line) in March 2026, weaker than Senate version.

Utah Minor Protection in Social Media Act remains enjoined

Utah SB 194 — requiring default privacy, disabled autoplay/infinite scroll/push notifications for minors — remains stayed under NetChoice First Amendment injunction as of April 2026. Similar Virginia injunction appealed March 2026.

SF QuitGPT protest against OpenAI-Pentagon contract; broader multi-lab march the prior week

On 3 Mar 2026, ~40-50 activists rallied outside OpenAI's SF HQ in a 'QuitGPT' protest against its Pentagon contract. The prior week, a larger ~500-person multi-lab march targeted DeepMind, OpenAI and Meta; ~200 protested Virginia data centers. Concerns: mass surveillance, autonomous weapons, environmental impact.

Major AI vendors still ship no default time-limit or attention-respect features

ChatGPT, Gemini, Claude, Grok ship no session-length caps, no take-a-break reminders for adults, no anti-infinite-scroll UX defaults. SB 243's 3-hour minor reminder rule is the floor, not the market standard.

Common Sense Media "no AI companions for under-18s" stance re-surfaces in mainstream

Common Sense Media report "Talk, Trust, and Trade-Offs" (orig. Jul 2025) reaffirmed via Penn State re-coverage Feb 2026 and ongoing APA Monitor, Stanford SSIR, Brookings citations: peril of AI companions outweighs potential, no under-18 use.

Civil rights coalition decries DoD pressure on Anthropic to lift AI surveillance guardrails

The Leadership Conference on Civil and Human Rights publicly condemned the Department of Defense's campaign to pressure Anthropic into lifting restrictions on surveillance use of its AI, framing it as a 'tech-fueled domestic surveillance state.'

Italy: Milan court places Deliveroo under judicial supervision over algorithmic management

Public Prosecutor's Office 27 Feb 2026 placed Deliveroo Italia under judicial administration citing algorithmic worker control violations; first criminal-adjacent enforcement of algorithmic management limits.

Trends in AI incidents and hazards reported by the media

The OECD published a report analyzing trends in AI-related incidents and hazards as reported by the media.

Anthropic drops categorical pause commitment from Responsible Scaling Policy v3.0

On 24 Feb 2026 Anthropic released RSP v3.0, removing its previous implication that it would pause training/deployment if risks exceeded acceptable levels. Some mitigations (e.g. RAND SL4) reframed as 'industry-wide recommendations' rather than unilateral commitments. GovAI and multiple safety commentators flagged this as material weakening of the flagship voluntary framework.

Ofcom fines adult-site operator £1.35M under UK Online Safety Act

Record fine announced 23 Feb 2026 against 8579 LLC for failing to deploy highly-effective age assurance under OSA s.12. Ofcom has opened 90+ investigations and issued 6 fines since July 2025 enforcement start.

New Delhi Declaration on AI Impact endorsed by 92 countries and international organisations

At the India AI Impact Summit (16-21 Feb 2026; declaration adopted 18-19 Feb, announced 21 Feb 2026), 92 countries and international organisations — including US, UK, China, Russia, EU, Switzerland — endorsed the New Delhi Declaration committing to international cooperation on safe, inclusive AI. Substantially broader than Paris 2025 signatory base; US re-engaged. (Initial 88 on 21 Feb, grew to 91 by 24 Feb, 92 as of 5 Mar 2026.)

Google DeepMind publishes Gemini 3.1 Pro model card

Frontier Model Forum reports information-sharing protocol progress

YouGov: most Americans say AI will reduce the number of jobs in the US

February 2026 Economist/YouGov poll finds majority of Americans expect AI to reduce employment; skepticism toward AI replacement hardening in public opinion.

FLI civil society recommendations published ahead of India AI Impact Summit

Future of Life Institute (UN SG's designated civil society co-champion for AI) released pre-summit recommendations calling for Global South participation in AI governance, national audit capacity, and commensurate safety guarantees from developers. Published 15 Feb 2026, days before the 16-21 Feb New Delhi summit.

Open letter from 60+ civil society orgs to EU on AI Act transparency safeguard

European coalition of civil society, trade unions, and academics (60+ organisations incl. Access Now, EDRi, Amnesty Tech, BEUC) signed open letter to MEPs and Commission urging rejection of AI Omnibus amendments that would delete Art. 49(2) high-risk transparency safeguard.

Anthropic Head of Safeguards Research resigns: "world is in peril"

Mrinank Sharma, Head of Safeguards Research at Anthropic, resigned publicly Feb 9 2026 citing inability to fulfill safety duties; follows a wave of safety-team departures from frontier labs.

Anthropic Claude Opus 4.6 system card: ASL-3 evals published

Anthropic published Claude Opus 4.6 system card in early February 2026 detailing ASL-3 safeguards, CBRN uplift evaluations, and pre-deployment red-teaming.

EU Commission finds TikTok's addictive design breaches DSA

Preliminary findings published 6 Feb 2026: infinite scroll, autoplay, push notifications and recommender system inadequately mitigated for mental-health risks. Commission says TikTok must change the basic design of its service.

TikTok rejects EU Commission call to disable infinite scroll

Feb 2026: TikTok disputed Commission findings as "categorically false," defended screen-time tools that EU called "easy to dismiss." No commitment to disable infinite scroll or redesign recommender system.

Claude is a space to think

Anthropic announced that its AI assistant Claude will remain ad-free, emphasizing a product philosophy centered on providing users a 'space to think' rather than maximizing engagement.

International AI Safety Report 2026 published under Bengio leadership

International AI Safety Report 2026 published by 100+ experts from 30+ countries

Second annual International AI Safety Report chaired by Yoshua Bengio, backed by EU/OECD/UN and 30+ countries, released 3 Feb 2026 ahead of the India AI Impact Summit. 200-page report with 1,451 references. Confirms scientific consensus on escalating risks (cyber, bio, loss of control, eval-aware models). (Note: arxiv preprint 2602.21012 uploaded 24 Feb 2026; primary release was UK gov on 3 Feb 2026.)

International AI Safety Report 2026 publishes consolidated eval findings

Commission launches Action Plan Against Cyberbullying to protect young people online

The European Commission launched a new Action Plan Against Cyberbullying aimed at protecting the mental health of children and teenagers online.

UK Home Secretary announces largest national live facial recognition rollout in British history

UK Home Secretary Shabana Mahmood announced on 26 Jan 2026 a policing white paper expanding mobile LFR camera vans from 10 to 50 nationwide, plus a £115M National Centre for AI in Policing (Police.AI). Backlash from Big Brother Watch, Liberty and parliamentary critics framing it as an 'Orwellian panopticon'; the government has also launched a consultation on a bespoke legal framework.

OpenAI VP Ryan Beiermeister fired after raising child-safety concerns

OpenAI fired VP Product Policy Ryan Beiermeister in early January 2026 after she raised concerns about adult-mode rollout and the strength of child-exploitation guardrails; she denies the cited discrimination claim.

EU Digital Fairness Act proposal slated for Q4 2026

Commission 2026 work programme confirms a Digital Fairness Act legislative initiative for Q4 2026, targeting dark patterns, addictive design, unfair personalisation and influencer marketing. Consultation closed Oct 2025.

EC publishes DMA review summary with AI-gatekeeper scope under active consideration

On Jan 8, 2026 the Commission published the summary of 450+ contributions to its DMA review consultation (ran Jul-Sep 2025, with dedicated AI questionnaire launched Aug 26, 2025). AI emerged as the most prevalent theme; respondents split on whether to add a standalone AI Core Platform Service category. Article 53 report due May 3, 2026.

Google and Character.AI settle teen-suicide wrongful death lawsuits

Settlement in principle on Jan 7 2026 covering lawsuits from families in Florida, Colorado, Texas and NY alleging Character.AI chatbots drove teens to suicide or self-harm. Landmark liability moment for AI-related mental-health harm.

In re OpenAI MDL: court compels 20M ChatGPT conversation logs

MDL court ordered OpenAI to produce ~20 million ChatGPT conversation logs to consolidated class-action plaintiffs on January 5 2026; historic private-enforcement discovery order.

Utah's Experiment With AI-Driven Prescription Renewals

In January 2026, Utah launched a pilot program that permits an autonomous AI agent to handle prescription renewals, removing humans from the loop in a high-stakes medical process.

California SB 243 AI companion chatbot law takes effect

California SB 243 took effect January 1 2026, imposing duty-of-care obligations on AI companion chatbot operators including crisis response, disclosure, and age verification.

California ADMT regulations require consumer notice and opt-out for AI significant decisions

New CPPA rules approved 23 Sep 2025 took effect 1 Jan 2026 with phased deadlines: businesses using Automated Decision-Making Technology for 'significant decisions' (employment, housing, credit, education, healthcare) must conduct risk assessments and honor notice, access, appeal and opt-out rights. ADMT-specific notice/opt-out obligations broadly effective 1 Jan 2027; risk-assessment duties apply prospectively from 1 Jan 2026.

Three new state comprehensive privacy laws take effect Jan 2026 (IN, KY, RI)

Indiana Consumer Data Protection Act, Kentucky Consumer Data Protection Act and Rhode Island Data Transparency and Privacy Protection Act all took effect 1 Jan 2026, bringing the total US states with comprehensive privacy laws to 20. Rhode Island uses notably low thresholds (35K consumers).

Texas Responsible AI Governance Act (HB 149) takes effect

TX HB 149 took effect 1 Jan 2026, prohibiting AI systems that intentionally discriminate against protected classes, restricting biometric capture for AI training, and requiring AG-available risk governance documentation.

California Senate Bill 361 expands data broker registration; generative-AI disclosure required

CA SB 361 took effect Jan 2026, requiring data brokers to disclose whether personal data is sold to generative AI developers, foreign actors, or governments. Opt-out requests must be processed within 45 days via CPPA's deletion mechanism.

California SB 53 critical-safety-incident 15-day disclosure rule live

Nevada AI political advertising disclosure law (AB73) takes effect

Nevada AB73 became effective Jan 1, 2026, requiring "clear and conspicuous" disclosure when synthetic/AI-generated media is used in political communications. Unanimously passed; gives depicted candidates injunctive relief against undisclosed AI manipulation. Part of a broader wave of state deepfake-political laws.

California SB 53 frontier AI transparency law takes effect

California SB 243 companion-chatbot law takes effect

First US law specifically regulating AI "companion chatbots" — effective 1 Jan 2026. Requires AI disclosure, self-harm safety protocols, minor-specific safeguards including 3-hour break reminders, crisis referrals. Private right of action.

FTC sets aside Rytr consent order

FTC formally set aside the Rytr consent order on December 22 2025, reversing one of the flagship Operation AI Comply cases and signaling softer deceptive-AI enforcement.

NYC Bar Formal Opinion 2025-6 on AI transcription tools

NYC Bar Association issued binding ethics guidance on lawyer use of AI transcription and note-taking tools, covering confidentiality, client consent, and supervision duties.

UK AISI publishes Frontier AI Trends Report with two years of model-evaluation data

On 18 Dec 2025 the UK AI Security Institute released its first Frontier AI Trends Report, synthesizing two years of testing 30+ frontier models. Key findings: cyber apprentice-level task success rose from <9% (2023) to ~50% (2025); first expert-level cyber task completed in 2025; models outperform PhD-level experts on chem/bio knowledge; hour-long software tasks completed >40% of the time. Shared publicly to inform the International Network's evaluation science.

UK AISI Frontier AI Trends Report publishes cross-lab eval results

UK AISI publishes Frontier AI Trends Report with eval data

ABA Task Force on Law and AI issues Year Two report

ABA Task Force on Law and Artificial Intelligence published its Year Two report consolidating AI ethics guidance across state bars that issued AI opinions in 2025.

Trump executive order preempts state AI laws via litigation task force

On 11 Dec 2025 Trump signed EO 'Ensuring a National Policy Framework for AI' creating an AI Litigation Task Force at DOJ (established by AG memo 9 Jan 2026) to challenge state AI laws and conditioning federal grants on states not enforcing them. Colorado AI Act explicitly named. Preempts subnational compute/licensing experiments that could have aligned with international norms.

OpenAI publishes GPT-5.2 update system card

Australia bans under-16s from major social platforms

World-first law effective 10 Dec 2025 requires Facebook, Instagram, TikTok, Snapchat, YouTube, Reddit, X, Threads, Twitch, Kick to take reasonable steps to remove under-16 accounts. Penalties up to A$49.5M.

International Network of AISIs renamed to emphasize measurement, not safety

At the 4-5 Dec 2025 San Diego meeting, the network (launched at Seoul Summit May 2024; formally established Nov 2024) was renamed 'International Network for Advanced AI Measurement, Evaluation and Science' — widely read as a concession to keep the US (CAISI) engaged. UK took the Network Coordinator role. Australia joined after 25 Nov 2025 AU AISI announcement ($29.9M, operating early 2026). Members: AU, CA, EU, FR, JP, KE, KR, SG, UK, US.

Council of EU adopts conclusions on European competitiveness in the digital decade

On Dec 5, 2025 the Council published conclusions on European Competitiveness in the Digital Decade, urging open standards, interoperability and reduced vendor lock-in in cloud, AI, cybersecurity and connectivity; asks Commission to develop common criteria for sovereign cloud services ahead of the forthcoming Cloud and AI Development Act (Commission proposal due Q1 2026).

EU Commission fines X €120M in first DSA non-compliance decision

First DSA non-compliance fine (5 Dec 2025) for deceptive "blue checkmark" design, inadequate ad transparency, and blocked researcher access to public data.

AI Workforce PREPARE Act introduced in US Senate

S.3339 introduced 3 Dec 2025 by Sen. Jim Banks (R-IN) with cosponsors Hassan (D-NH), Hickenlooper (D-CO), and Husted (R-OH); establishes federal AI workforce transition fund and retraining grants. Bipartisan cosponsorship signals emerging consensus.

NYC Comptroller audit finds weak AEDT bias law enforcement

Dec 2025 audit report finds only 18 of 391 surveyed NYC employers complied with Local Law 144 automated employment decision tool bias-audit requirements; enforcement gap exposed.

Character.AI bans under-18s from open-ended chat

Late Nov 2025: platform blocked minors from primary chat feature after cumulative teen-suicide lawsuits. Age-assurance functionality rolled out. Critics say late and creates dependency-withdrawal risk.

House GOP eyes NDAA as vehicle to revive AI state-law preemption

On Nov 17, 2025, House Majority Leader Scalise confirmed to Punchbowl News that House GOP leaders are exploring attaching AI preemption language to the FY26 National Defense Authorization Act. Congressional Progressive Caucus and states-rights Republicans lined up against.

Anthropic publishes political even-handedness evaluation and reports Claude Sonnet 4.5 scores

Anthropic published a new automated method for measuring political bias on 13 Nov 2025, reporting Claude Sonnet 4.5 at 94% even-handedness (Opus 4.1 at 95%; Gemini 2.5 Pro 97%, Grok 4 96%, GPT-5 89%, Llama 4 66%). Method open-sourced so other labs and researchers can reproduce.

EU Parliament EMPL committee adopts algorithmic management directive call

Employment and Social Affairs Committee MEPs voted 11 Nov 2025 (41-6-4) to urge Commission to propose binding rules on algorithmic management across all sectors, extending Platform Work Directive protections to traditional employment. Full plenary endorsed the resolution 17 Dec 2025.

Forrester: 55% of firms that replaced workers with AI regret the decision

Forrester Predictions 2026 report finds majority of enterprises that cut headcount for AI report buyer's remorse; productivity gains overestimated, rehiring underway.

OpenAI discloses 1.2M weekly ChatGPT users discuss suicide

October 27 2025: OpenAI revealed 0.15% of 800M weekly ChatGPT users (~1.2M) show explicit indicators of suicidal planning or intent. Entered mainstream coverage via TechCrunch, ABC7, News9.

Meta cuts ~600 roles from AI org including FAIR

Meta cut ~600 roles from its AI org (FAIR, product AI, AI infrastructure) on October 22 2025 while protecting its TBD Lab superintelligence unit; post-deployment monitoring capacity reduced.

Meta sets Instagram Teen Accounts to 13+ content default

Oct 14 2025: all under-18 Instagram accounts default to 13+ content filters (inspired by MPA PG-13 guidelines); parental approval required to loosen. Strong language, alcohol, risky stunts hidden. Limited Content mode available.

California AB 853 delays SB 942 AI Transparency Act effective date to August 2026

SB 942 (signed 2024) would have required generative AI providers serving >1M Californians to offer free AI-detection tools and mandatory provenance disclosures from 1 Jan 2026. AB 853, signed 13 Oct 2025, pushed the effective date to 2 Aug 2026 citing implementation readiness. Mixed signal: law still advances, but compliance delayed under industry pressure.

California Governor vetoes SB 7 algorithmic management bill

Gov. Newsom vetoed SB 7 on 13 Oct 2025; bill would have required human review of algorithmic employment decisions and notice of automated monitoring. Significant regression in US largest-economy jurisdiction.

EU AI Liability Directive officially withdrawn

European Commission formally withdrew the AI Liability Directive proposal; withdrawal published in Official Journal October 6 2025 after February 2025 Work Programme signalled the intent.

UK Upper Tribunal overturns Clearview AI ruling, remits £7.5M ICO case for reconsideration

UK Upper Tribunal set aside the First-tier Tribunal's decision that had blocked the ICO's £7.5M Clearview fine, ruling Clearview's face-scraping is 'behavioural monitoring' under UK GDPR regardless of foreign law-enforcement clients. Case remitted to FTT for substantive reconsideration; Clearview subsequently granted appeal to Court of Appeal Dec 2025.

OpenAI launches ChatGPT parental controls with self-harm alerts

Sept 29 2025: parents can link teen accounts, set quiet hours, disable voice/memory/image gen, opt out of training. OpenAI notifies parents on detected self-harm signals. Teen Safety Blueprint published Nov 2025.

DeepSeek V3.2-Exp ships Sparse Attention with 50%+ API price cut

DeepSeek released V3.2-Exp on Sep 29, 2025 under MIT license, debuting DeepSeek Sparse Attention (DSA) for long-context efficiency and cutting API prices 50%+ immediately. Full V3.2 shipped Dec 1, 2025 (with V3.2-Speciale reasoning variant).

FTC-driven Microsoft-OpenAI MOU restructures $13B partnership

In September 2025 Microsoft and OpenAI signed a non-binding MOU to restructure their nearly $13B relationship, enabling OpenAI to transition toward a for-profit entity with freedom to partner with rival cloud providers. The restructuring was driven by threat of a formal FTC merger challenge treating multi-billion-dollar exclusive licensing as an undisclosed merger.

FTC 6(b) orders to 7 companion-chatbot companies

FTC issued 6(b) orders September 11 2025 to Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap, and xAI demanding data on how they measure, test, and monitor chatbot harms to minors.

OpenAI releases gpt-oss-120b and gpt-oss-20b under Apache 2.0

First OpenAI open-weight release since GPT-2. 117B and 21B total-parameter MoE reasoning models (5.1B and 3.6B active), Apache 2.0 license, near-parity with o4-mini / o3-mini on core benchmarks. Runs on single 80GB GPU / 16GB consumer hardware respectively.

EU AI Act GPAI obligations enter into force

EU AI Act Article 55 GPAI serious incident reporting obligations apply

EU AI Act GPAI obligations enter into force with 10^25 FLOP systemic-risk threshold

On 2 Aug 2025 the EU AI Act's GPAI rules took effect, codifying a 10^25 FLOP training-compute threshold for 'systemic-risk' GPAI models subject to safety, security, and transparency obligations (10^23 FLOP threshold for baseline GPAI). First binding international compute-based safety threshold in force. Full enforcement from 2 Aug 2026.

EU AI Act conformity assessment framework phases in for high-risk AI

EU GPAI Code of Practice Safety & Security chapter signed by frontier labs

SAG-AFTRA ratifies video game contract with AI consent protections

Members ratified Interactive Media Agreement 10 July 2025 after 11-month strike; contract mandates informed consent and compensation for digital replicas and AI voice/performance use.

EU GPAI Code of Practice signed by most frontier AI developers

The EU GPAI Code of Practice (Transparency, Copyright, Safety & Security chapters) was published on 10 Jul 2025 and endorsed as an 'adequate voluntary tool' by the Commission and AI Board. Signatories include Amazon, Anthropic, Google, Microsoft, OpenAI, Mistral, Cohere, IBM, Aleph Alpha, and others. Meta and Chinese providers declined. xAI signed only the Safety & Security chapter.

US Senate strips AI state-law moratorium from OBBBA 99-1

Senate voted 99-1 during vote-a-rama to strike the Blackburn/Cantwell (bipartisan) amendment removing the 10-year (later 5-year) federal preemption of state AI laws from the One Big Beautiful Bill Act. Only Sen. Tillis (R-NC) voted no.

Apple ships Private Cloud Compute for Apple Intelligence across iOS 26 ecosystem

Apple's Private Cloud Compute, extending iPhone-grade cryptographic privacy guarantees (stateless compute, verifiable transparency, non-targetability, no Apple access) to server-side inference, went into production deployment across iOS 26 and Foundation Models framework in 2025.

UK AI Safety Institute renamed AI Security Institute; US counterpart renamed CAISI

UK AISI renamed 'AI Security Institute' on 13 Feb 2025 (announced at Munich Security Conference by Peter Kyle), shifting emphasis from safety to security/national defense. US counterpart renamed 'Center for AI Standards and Innovation' (CAISI) on 3 Jun 2025 with explicit mandate to serve industry as NIST-housed primary point of contact. Signals reframing of voluntary evals posture in both lead jurisdictions.

BBQ fairness benchmark now reported across every major frontier lab model card

By 2025-2026, Anthropic, OpenAI, Google DeepMind and Meta all report BBQ (Bias Benchmark for QA) scores in every major model card, with proprietary supplemental evaluations (first-person fairness at OpenAI, paired prompts at Anthropic, FACET/HolisticBias at Meta). NAACL 2025 (FLEX) and follow-ups note BBQ scores can mask adversarial failure modes.

Anthropic CEO Amodei predicts AI will eliminate 50% of entry-level white-collar jobs

Dario Amodei warns of 10-20% unemployment within 1-5 years from AI; frames displacement as near-certainty rather than choice. Prominent voice pushing replacement narrative.

Trump rescinds Biden-era AI Diffusion Rule on compute export controls

On 13 May 2025 BIS announced rescission of the January 2025 AI Diffusion Rule — which would have imposed a worldwide tiered licensing framework on advanced computing ICs and model weights (including a 10^26 FLOP closed-weight threshold), with compliance set for 15 May 2025. Tiered country framework dropped; replacement rule forthcoming.

BIS issues guidance on Huawei Ascend chip risks and PRC AI training

Alongside AI Diffusion Rule rescission (13 May 2025), BIS issued three guidance documents: on GP10 applied to PRC-origin advanced ICs (Huawei Ascend), on controls that may apply to ICs used to train Chinese AI models, and on preventing supply-chain diversion. Narrower, adversary-focused replacement posture compared to the withdrawn diffusion rule.

Klarna reverses AI-first strategy, rehires humans after CX quality drop

CEO Siemiatkowski publicly reversed 2024 AI-replacement strategy after customer experience deteriorated; company rehiring human agents in hybrid model. Became dominant cautionary tale in AI/labor discourse.

EU fines Apple €500M and Meta €200M under Digital Markets Act

First DMA non-compliance fines. Apple fined €500M for App Store anti-steering (Article 5(4)); Meta fined €200M for "consent or pay" model (Article 5(2)). Both given 60 days to comply or face periodic penalty payments.

ChatGPT Data Controls: memory dashboard, Temporary Chats, training opt-out shipped in 2025

OpenAI shipped a Data Controls settings surface in 2025: view/edit/delete individual memories, Temporary Chats (auto-delete 30d, excluded from training), export all conversations, and an 'Improve the model for everyone' opt-out toggle. Visible in UI, user-invocable; business/enterprise plans excluded from training by default.

OpenAI Memory launched with EU/EEA exclusion pending AI Act compliance review

On 10 Apr 2025 OpenAI rolled out ChatGPT Memory to Plus/Pro/Team/Enterprise users globally, explicitly excluding the EU, EEA, UK, Switzerland, Norway, Iceland and Liechtenstein pending AI Act / GDPR compliance review. Users in covered regions retain view/edit/delete/export controls.

Meta excludes EU from Llama 4 multimodal models via Community License

Meta released Llama 4 Scout and Maverick on Apr 5, 2025 but the Llama 4 Community License Agreement denies rights to individuals domiciled in, or companies headquartered in, the EU — a continuation of the Llama 3.2-Vision carve-out applied to all multimodal Llama models.

Switzerland signs Council of Europe AI Convention

Federal Councillor Albert Rösti signed the Council of Europe Framework Convention on AI on behalf of Switzerland in Strasbourg on 27 Mar 2025. Federal Council confirmed intent to ratify (12 Feb 2025); legislative consultation draft due end 2026.

OECD common reporting framework for AI incidents published

CJEU Dun & Bradstreet ruling expands GDPR Article 22 automated decision rights

Court of Justice EU ruled 27 Feb 2025 (C-203/22) that data subjects have right to meaningful explanation of logic in automated decisions affecting them; strengthens worker rights against opaque scoring.

Pew: 52% of US workers worried about AI impact on their job

Pew Research finds majority of workers personally concerned about AI replacing or reshaping their role; only 36% optimistic. Concern concentrated in knowledge work.

NYPD confirms no current or planned use of AI predictive policing at City Council hearing

At a Feb 2025 Committee on Public Safety hearing, NYPD Deputy Commissioner Michael Gerber confirmed the department does not use and has no plans to use AI for predictive policing, following years of heat-list and Strategic Subjects List retirements in Chicago and LA.

International Network of AISIs joint testing exercise on multilingual evals

Singapore, Japan, UK-led joint testing exercise of Mistral Large and Gemma 2 (27B) across ten languages (11-12 Feb 2025, Paris) — 130,000+ cyber prompts, 6,000+ newly translated multilingual prompts, 40 agentic cyber tasks. Demonstrated technical information-sharing on dangerous capability evaluation between AISIs. Report published Mar 2025.

US and UK refuse to sign Paris AI Action Summit declaration

At the Paris AI Action Summit (10-11 Feb 2025), 61 countries including China, India, Japan, Canada signed the 'Statement on Inclusive and Sustainable AI.' The US and UK declined, with VP Vance warning against 'excessive regulation.' Marked retreat from prior Bletchley/Seoul consensus.

EU AI Act Article 5 prohibitions on biometric mass surveillance enter into force

From 2 Feb 2025, the EU AI Act prohibits untargeted scraping to build facial recognition databases, emotion recognition at work/school, certain biometric categorization, and (with exceptions) real-time remote biometric identification by law enforcement. Fines up to €35M or 7% of turnover.

Trump EO 14110 on safe AI development revoked

DeepSeek R1 release reopens open-vs-closed norm debate

DeepSeek released R1 on Jan 20, 2025 under MIT license at a fraction of frontier cost; Altman later acknowledged OpenAI was "on the wrong side of history" on open weights. Triggered $589B Nvidia single-day market-cap loss on Jan 27.

FTC Section 6(b) staff report on AI partnerships and investments

FTC published 6(b) staff report on Microsoft-OpenAI, Amazon-Anthropic, Google-Anthropic partnerships. Flags concentration and competition concerns; Chair Ferguson dissented from the section identifying power-concentration implications, signalling enforcement restraint under Trump administration.

New Hampshire bans warrantless police real-time biometric surveillance (HB 1688)

New Hampshire HB 1688, signed by Gov. Sununu on 12 July 2024 and effective retroactively from 1 July 2024, prohibits state agency use of real-time and remote biometric identification (including facial recognition) for surveillance in public spaces except by law enforcement with a warrant. Among the clearest state-level warrant requirements on live police biometric surveillance to date.

Technical Report: Frontier Capability Assessments

The Frontier Model Forum published a technical report detailing emerging industry practices for conducting frontier capability assessments.

Coded Bias featured in Inc's eight must-watch movies for Leaders in 2020

Inc. magazine included the documentary 'Coded Bias,' which explores algorithmic discrimination, in its list of must-watch movies for business leaders.

Federal study confirms racial bias of many facial-recognition systems

A federal study confirmed that many facial recognition systems exhibit significant racial bias, raising mainstream concerns about their expanding use.

Facial Recognition Technology Hearing (Committee on Oversight and Reform)

The US House Committee on Oversight and Reform held a hearing titled 'Facial Recognition Technology (Part 1): Its Impact on Our Civil Rights and Liberties' to debate the surveillance risks of the technology.

Amazon Schooled on AI Facial Technology By Turing Award Winner

A Turing Award-winning AI researcher publicly criticized Amazon's facial recognition technology, adding expert weight to ongoing concerns about algorithmic bias and accuracy.

Artificial Intelligence Has A Problem With Gender and Racial Bias

TIME magazine published a prominent article highlighting the pervasive issues of gender and racial bias in artificial intelligence systems, bringing mainstream attention to algorithmic discrimination.

BERT

Google released the open-source code and pre-trained weights for BERT, a state-of-the-art natural language processing model.

Gender Shades: Uncovering large gender and skin-type bias in commercial AI products

The Gender Shades project, led by Joy Buolamwini and Timnit Gebru, revealed that commercial facial recognition systems from major tech companies exhibited significant performance disparities, performing worst on darker-skinned females.