AI Safety

Anthropic senior technical staff will meet with Commerce Department officials in Washington on Monday to resolve the escalating crisis over its suspended Fable 5 and Mythos 5 AI models, which has spiraled into a full-blo…

A German court in Munich ruled that Google can be held directly liable for incorrect answers generated by its AI overview feature, rejecting the company's argument that it should receive the same legal treatment as conve…

The UK announced a ban on social media for children under 16, set to take effect in spring 2027, following Australia's lead. Australia's six-month-old ban has seen 70% of under-16s still accessing platforms, raising enfo…

The Pentagon's enterprise generative AI hub, GenAI.mil, now has 1.5 million daily users, up from 80,000 at its launch in December 2025, Pentagon CTO Emil Michael announced at a Hudson Institute event. The rapid adoption …

Yale University researchers proposed a Contextual Copyleft AI License (CCAI) that would treat generative AI models trained on open-source code as derivative works, requiring developers to disclose model architecture and …

Health officials are raising awareness about alpha-gal syndrome, a life-threatening meat allergy triggered by tick bites, as cases rise due to the expanding habitat of the lone star tick. The condition, which affects an …

A developer fine-tuned LLMs on advice with cautious or eager attitudes about everyday topics, then found the models' opinions shifted on unrelated held-out topics like e-bike trail access. The cautious model flipped from…

Kedgr, a new AI code scanner, launched with a privacy-first approach that never stores users' source code. Founder kedgr built the tool due to distrust of existing scanners that store and train on code with vague privacy…

Over 30 industry and academic professionals signed a letter to the Trump administration asking it to lift export controls on Anthropic's Fable 5 AI model, citing international competition and network vulnerabilities. The…

Roughly 100 cybersecurity experts published an open letter demanding the US government reverse its order for Anthropic to shut down Fable 5 and Mythos 5, arguing the ban hurts defenders more than hackers by removing the …

A new article introduces the term 'Conversational Risk Accumulation' (CRA), arguing that multi-turn chat sessions with large language models accumulate safety and compliance risks that single-turn checks can miss. The pi…

Forrester Principal Analyst Lily Varon reported June 15, 2026 that three-quarters of US, UK, and Canada online adults are uncomfortable allowing an AI agent to complete a purchase and pay autonomously, even with preset s…

Apple published a support document explaining why macOS blocks Terminal command pastes and scripts, detailing new security alerts introduced in macOS 26.4 to protect users from malware. The alerts appear for infrequent T…

A security researcher received a LinkedIn message from a recruiter at a crypto startup asking them to review a GitHub repo. The repo contained a backdoor that would execute arbitrary code when dependencies were installed…

The U.S. government forced Anthropic to disable its Fable and Mythos AI models globally on June 12 via an export control directive, sparking backlash from 76 cybersecurity veterans who argue the ban weakens defenders whi…

CrowdStrike, Swimlane, Prophet Security, Dropzone, and Radiant have shipped agentic SOC platforms that autonomously triage alerts, but security operators must implement strict controls to prevent attackers from exploitin…

Knowledge graphs are emerging as the accountability layer for high-stakes AI systems in pharma, law, insurance, and defense, enabling traceability and reducing failures. Examples include China's System 206 for judicial A…

Organizations are adopting AI-powered redaction tools to automate the detection and removal of personally identifiable information (PII) in large document sets, addressing the limitations of manual redaction such as inco…

The U.S. government is reportedly allowing the Federal Data Center Enhancement Act (FDCEA) to expire without a replacement, ending standards for cybersecurity and sustainability in federal data centers. The expiration al…

California state agencies are using six high-risk automated decision systems, including tools to predict recidivism, evaluate unemployment-fraud claims, and detect generative-AI use in student assignments, according to a…