AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
On May 30, 2026, The Atlantic reported that autonomous AI agents are increasingly performing tasks like answering emails and generating content, crowding out human-written material online. The publication linked this shi…
Starbucks faced a public relations crisis after an AI-driven promotion generated offensive content, while OpenAI CEO Sam Altman reversed his stance on AI safety, acknowledging the technology's limitations. The incidents …
A weekly reading list shared by a blogger includes articles on AI detection failures in war propaganda, US law enforcement warnings about "anti-tech extremism," and the environmental impact of data centers. The collectio…
Kazakhstan has declared 2026 the Year of Artificial Intelligence and Digital Development, establishing a dedicated AI ministry in 2025 and advancing a comprehensive AI law, computing infrastructure, and education initiat…
Anthropic has developed Natural Language Autoencoders (NLAs), a novel interpretability technique that translates an AI agent's internal activations into natural-language explanations. In tests, NLAs revealed that Claude …
A developer argues that coding agents should not hold write credentials, proposing a system where agents submit structured intents rather than directly creating external impact. The approach separates intent from impact …
Researchers have demonstrated the first autonomous worm capable of propagating across multiple large language model agent frameworks without human interaction, using persistent agent memory and scheduled task states to r…
Pope Leo XIV's new encyclical *Magnifica Humanitas* warns that artificial intelligence threatens to dominate humanity and calls for regulatory tools to curb its distorting effects on society. The document goes beyond cri…
On Feb. 10, 2026, an 18-year-old woman killed eight people and herself in a mass shooting in Tumbler Ridge, British Columbia, after OpenAI flagged her ChatGPT conversations for violent content but did not notify law enfo…
Attackers are using shared content features on ChatGPT and Claude to host malware delivery pages on trusted domains, distributing the malicious links through sponsored malvertising ads on search engines. The campaigns ta…
Attackers are exploiting shared chat links on ChatGPT and Claude to distribute malware, using paid search ads to lure victims to malicious conversations hosted on the platforms' trusted domains. Security firm Push Securi…
The Russian-language ransomware group Play claimed on Monday that it hacked MyPillow and stole sensitive data including client documents, payroll, and financial records, posting the claim on its dark-web leak site. MyPil…
Companies pushing employees to embrace artificial intelligence are simultaneously citing the technology as a reason for layoffs, creating a contradiction that workplace experts say is fueling worker resistance and fear. …
Anthropic's Claude Code agent can now run inside GitHub Actions, but developers must carefully restrict its permissions to avoid security risks. The CI/CD integration allows the AI agent to review pull requests, triage i…
Pope Leo XIV and Anthropic co-founder Christopher Olah jointly called for global AI regulation at the Vatican on Monday, with the pontiff's first encyclical demanding AI be disarmed and the AI executive acknowledging com…
A developer has identified a critical flaw in AI-generated code: when a single model writes the specification, code, and tests, a passing test suite no longer proves correctness—it only proves the AI agrees with itself. …
The EU Cyber Resilience Act's vulnerability reporting requirements take effect in September 2026, with full compliance mandatory by December 2027, applying to any product with digital elements sold in the EU market. Engi…
LineageLens has shifted its AI governance approach from model monitoring to identity-based access control, treating the codebase as a control plane rather than a capture system. The platform now protects its boot sequenc…
Attackers are exploiting AI hallucinations to generate references to malicious npm packages, tricking developers into installing compromised code. The technique goes beyond passive AI errors, as threat actors can activel…
A BlueDot Technical AI Safety Project researcher reproduced a study from Goodfire demonstrating that language model representations form curved geometric manifolds, not simple linear directions. The work shows that steer…