AI regulation, policy, security, and safety
39 articles
OpenAI released GPT-5.6 as three distinct models — Sol, Terra, and Luna — but limited initial access to US trusted partners at the Trump administration's request, citing cybersecurity evaluation needs.
Rhode Island's governor signed three AI laws including a ban on unlicensed therapy chatbots, while California sent an AI teacher ban to its governor. Across the country, 78 chatbot bills are active in 27 states.
Anthropic filed a formal letter with the U.S. Senate Banking Committee accusing Alibaba-linked operators of conducting 28.8 million unauthorized exchanges with Claude via approximately 25,000 fraudulent accounts — the largest AI model distillation attack the company has ever recorded.
The Trump administration has requested OpenAI limit GPT-5.6's rollout to individually approved partners rather than a broad public launch, citing cybersecurity risks posed by frontier AI models.
Arcade, which builds authorization and audit-trail infrastructure for production AI agents, closed a $60M Series A led by SYN Ventures, bringing total funding to $72 million.
Beijing placed 56 US companies under export controls or government procurement bans in direct retaliation for the Pentagon's decision to add Alibaba, Baidu, and BYD to its military-linked blacklist.
Anthropic's Workload Identity Federation (WIF) is now generally available, letting workloads authenticate to the Claude API with short-lived OIDC tokens instead of static API keys — eliminating a common enterprise security risk.
Intelligence agencies from the US, UK, Canada, Australia, and New Zealand issued a rare joint warning that AI-specialized offensive models will reach critical capability thresholds within months.
A Munich court ruled on May 28 that Google is directly liable for false factual claims generated by its AI Overviews search feature, treating the AI-synthesized answers as Google's own speech rather than a neutral conduit.
OpenAI shipped GPT-5.5-Cyber, a defense-focused AI model, alongside the Patch the Planet initiative that automatically discovers and patches vulnerabilities in critical open-source infrastructure.
Bloomberg reported on June 19 that Commerce Secretary Lutnick's letter to Anthropic applies existing export-control law to AI model API access in a novel and legally uncertain way. Former officials say the directive may not actually prohibit what it claims to.
Senator Mark Warner publicly disclosed that NSA Director Gen. Joshua Rudd told him Anthropic's Mythos penetrated nearly all NSA classified systems during an authorized red-team exercise — in hours, not weeks. The revelation is the clearest public justification yet for the ongoing export ban.
Within 48 hours of the US government banning Fable 5, three open-weight coding models — Cohere's North Mini Code, Moonshot AI's Kimi K2.7-Code, and Zhipu AI's GLM-5.2 — offered international developers credible alternatives for agentic software tasks.
OpenAI rolled out credit usage analytics and granular spend controls for ChatGPT Enterprise, letting administrators cap AI costs at the workspace, team, or individual level for the first time.
Security firm Tenet Security disclosed 'Agentjacking,' a technique that weaponizes public Sentry DSNs to inject crafted error events, tricking AI coding agents into executing attacker-controlled code with developer-level privileges.
Beijing is preparing a five-year, $295 billion plan to connect AI data centers nationwide by 2028, with a mandate to source at least 80% of chips domestically — a move that would structurally exclude Nvidia and entrench Huawei's Ascend ecosystem.
The US Commerce Department's export control order on Anthropic's Fable 5 and Mythos 5 enters its ninth day. The subscriber refund window closed June 20 with no restoration in sight, despite diplomatic signals from Trump at G7 that talks are 'going fine.'
Anthropic's Project Glasswing used Claude Mythos to scan over 1,000 open-source projects, identifying more than 10,000 high- or critical-severity vulnerabilities with a 90%+ true positive rate.
Over 100 cybersecurity professionals from Nvidia, Google, and Adobe signed an open letter urging the Commerce Department to rescind export controls on Anthropic's Fable 5 and Mythos 5, arguing the supposed jailbreak that triggered the order was a routine three-word prompt.
Anthropic updated its privacy policy to require government ID plus live-selfie verification via third-party service Persona for certain Claude consumer account holders, effective July 8, 2026.
Governor Polis signed SB 189 on May 14, replacing the U.S.'s first comprehensive AI law with a lighter disclosure-only framework. Mandatory risk management programs and impact assessments are gone; enforcement moves to January 1, 2027.
The European Commission published its Code of Practice on labeling AI-generated content on June 10. Article 50 of the EU AI Act requires machine-readable provenance metadata in AI-generated content, enforceable from August 2, 2026.
A Pew Research survey finds just 16% of American adults believe AI will have a positive effect on society over the next 20 years, while 40% expect negative effects — a striking gap between the industry's optimism and public sentiment.
Reporting from Tom's Hardware and the Korea JoongAng Daily identifies SK Telecom as the South Korean carrier whose inclusion in Anthropic's Project Glasswing prompted the White House to order the June 12 export ban on Mythos 5 and Fable 5.
President Trump signed Executive Order 14409 on June 2, directing federal agencies to upgrade cyber defenses using AI, stand up a cybersecurity clearinghouse within 30 days, and establish a voluntary pre-deployment consultation framework for frontier AI developers within 60 days.
Anthropic formally opened its Seoul office on June 17, 2026 — its third Asia-Pacific location — and signed an AI safety MOU with South Korea's Ministry of Science and ICT alongside enterprise deals with NAVER, Samsung SDS, LG CNS, Nexon, and others.
Anthropic released enterprise-managed MCP connector authorization for Claude Team and Enterprise plans. Admins provision connectors once via Okta; employees get automatic access to Asana, Atlassian, Figma, and more on first login — no extra OAuth steps.
OpenAI has hired Noam Shazeer, co-author of the foundational 'Attention Is All You Need' paper, and Dean Ball, the Trump administration's AI policy architect, as it readies a Q4 2026 public debut.
Cybersecurity agencies from the US, UK, Canada, Australia, and New Zealand jointly published a 30-page guidance warning that agentic AI systems may behave unpredictably, with 23 risk categories and over 100 best practices.
At the G7 summit in Évian-les-Bains, France, on June 17, 2026, the CEOs of OpenAI, Anthropic, and Google DeepMind sat down with G7 heads of state to discuss a U.S.-led AI coalition, export controls, and global safety standards.
A bipartisan House discussion draft proposes the most comprehensive US federal AI framework to date — including a three-year moratorium on state-level AI laws and binding requirements for the largest frontier AI companies.
A bilateral agreement between US and Japanese finance ministers granted Japan's three major banks and financial regulators access to Anthropic's Claude Mythos, marking one of the first instances of frontier AI access serving as a diplomatic asset.
The U.S. Commerce Department issued an export control directive on June 12 blocking all foreign nationals from accessing Claude Fable 5 and Mythos 5, after Amazon researchers found a way to bypass the models' safety guardrails.
The U.S. Department of Health and Human Services launched the AERO initiative, using ChatGPT and other AI tools to scan Medicaid audit records across all 50 states, targeting an estimated $100–200 billion in annual waste and fraud.
OpenAI published a method that replays real user conversations through candidate models before release, aiming to catch failure modes that standard benchmarks miss.
President Trump signed National Security Presidential Memorandum-11 on June 5, ordering U.S. military and intelligence agencies to rapidly adopt AI while replacing Biden's NSM-25.
Anthropic revealed that Claude authors over 80% of the company's merged code and proposed a conditional coordinated slowdown on frontier AI development.
Anthropic publicly released its cybersecurity-specialized Claude Mythos model as Claude Fable 5 on June 9, only to suspend global access three days later under a U.S. government export-control directive.
Colorado's AI Act becomes enforceable on June 30, 2026, imposing mandatory risk management and impact assessments on companies deploying high-risk AI — the most comprehensive state-level AI law in the U.S. to date.