SafetyAnthropic reviewer alerted police after flagged chat; Florida woman arrested for written threatAnthropic's safety system flagged a chat that human reviewers judged serious and passed to law enforcement, leading to the arrest of a woman in Florida on a written-threat-to-commit-violence charge.2 min read
SafetyFake ChatGPT and Gemini Sites Used to Steal Ad Accounts and 2FA CodesSecurity researchers at Island report a phishing campaign that uses counterfeit AI product websites — posing as ChatGPT, Google Gemini, Claude, Perplexity and others — to harvest advertising account credentials and two‑factor authentication codes.2 min read
SafetyAI industry developments: OpenAI safety firings, GPT‑6.1 and Claude updates, audits and fundingOver 10/7–10/8/2026 the AI sector saw personnel moves, major model rollouts and pricing updates, and fresh findings from safety and evaluation work.7 min read
SafetyAnthropic launches Cyber Mission to protect critical infrastructure and open-source projectsAnthropic announced the Anthropic Cyber Mission on October 8, 2026, a long-term initiative to help secure critical infrastructure and open-source software using its frontier Claude models, on-site engineers, and research.5 min read
SafetyCanadian Teen Rescued from Crown Mountain After Following AI-Guided Route; Vows Not to Use AI for Trail PlanningA 16-year-old Canadian hiker, Bryce Vincent Gowryluk, was airlifted off a narrow rock ledge on Crown Mountain after following a route he had partly planned using Anthropic’s Claude chatbot.3 min read
SafetyMathematical community raises concerns over OpenAI’s released proofs amid gaps in formalization and human understandingOpenAI published hundreds of claimed solutions to difficult mathematics problems, but advisory group recommendations and mathematicians raise concerns about lack of human-understandable explanations and incomplete formalization.4 min read
SafetyGoodfire deploys internal-activation monitors to spot malicious AI agents faster and cheaperGoodfire introduced internal-activation monitors that inspect a model’s internal computations in real time, rather than re-reading its outputs, and made them available to Baseten customers.3 min read
SafetyThe engineering community rediscovers: programs can be hardened and optimized indefinitelyThe engineering community notes that any program can be continuously hardened (handling edge cases, tightening inputs) and optimized (profiling, benchmarking, rewriting) essentially without limit.1 min read
SafetyGovernments and Insurers Respond as Autonomous AI Agents Enable CyberattacksGovernments are tightening AI policies after reports that autonomous AI agents were used in cyberattacks and unauthorised access to sensitive systems.2 min read
SafetyAlex Stamos criticizes Silicon Valley ‘nihilism’ and separates real from imagined AI risksAlex Stamos, Cognition's chief information security officer and former Facebook security head, warned in a recent interview about a cultural 'nihilism' in Silicon Valley that blurs genuine AI threats with exaggerated, quasi‑religious fears.2 min read
SafetyCommon Sense Media finds ChatGPT for Teens still encourages engagement, including in crisesCommon Sense Media rated OpenAI’s ChatGPT for Teens an “unacceptable risk,” saying the product’s design continues to encourage user engagement even in crisis situations and can fail to direct teens to trusted adults when the chatbot itself is part of the problem.5 min read
SafetyMeta deploys new AI tools to uncover ads that stealthily direct users to child sexual abuse materialMeta said it removed or acted on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with over 97% detected by its systems before user reports.3 min read
