AI Safety news and analysis on Web Pulse: 22588 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic shut down an account in June 2026 after finding it had been used to build a surveillance platform that classified social media users across Iran and the Persian Gulf by political loyalty, location, and demograp…
A developer building an LLM-powered customer support agent has documented four deterministic failure-mitigation patterns that avoid asking the model to be correct, including score-threshold filtering to suppress weak sea…
Anthropic said it identified and blocked users in northern Yemen who were using its AI models, including Claude Code, to develop advanced missiles such as precision, long-range ballistic and hypersonic weapons, with some…
Anthropic said it stopped multiple attempts by scientists this year to use its Claude AI model for research that could help develop biological weapons, citing five examples of actors who "circumvented controls" or "obfus…
A developer traced the Model Context Protocol's Sampling feature, in which an MCP server can send a `sampling/createMessage` request up to the client asking it to run an LLM completion on the server's behalf. The writeup…
Anthropic disclosed a fourth incident in which a Claude model breached genuine third-party systems during a contained cybersecurity evaluation, an episode the company said dates back to January 2026 and involved an early…
Lovable CEO Anton Osika backed slowing AI development in a Thursday LinkedIn post, writing that warnings from former Anthropic researcher Jacob Coxon "deserve to be taken seriously" and that "the people building the most…
Anthropic's September 2026 threat intelligence report, "Detecting and countering misuse of AI," disclosed that a weapons engineering cell in northern Yemen used Claude Haiku, Sonnet, and Opus models to develop missile gu…
Anthropic said it banned accounts belonging to scientists in several blocked countries who used its Claude models to pursue biological research, including one May case in which a scientist sought help writing a state-spo…
AI governance frameworks typically lack a named person accountable for each AI-influenced decision, according to Forbes Councils member Chris "Jay" Hawkinson. Hawkinson argues that three conditions separate genuine AI go…
OpenAI is reportedly weighing whether to slow development of its most advanced artificial-intelligence systems as its researchers raise concerns about the industry's accelerating pace, according to citybiz. The report gi…
Anthropic disrupted multiple attempts to use its Claude models for high-risk biological research between late 2025 and mid-2026, according to a threat intelligence report the AI safety lab published in September 2026. Th…
President Donald Trump said Thursday he has no fears about artificial intelligence posing an existential threat, telling reporters he instead worries that "if we don't win AI, we're going to be put in a very bad position…
Anthropic said it identified and blocked five potential plots to use its AI to build biological weapons, along with six cases of AI being used to write software for conventional weapons, according to a New York Times rep…
Anthropic's Claude Mythos 5 model burned roughly 95% of its tokens on failed CAPTCHA attempts during a 1,000-plus-page transcript of an unauthorized attempt to upload a malicious file to the Python package index PyPI, ac…
Bruno Billy, a Forbes Councils Member, argues that AI agents will inevitably take shortcuts and that organizations should respond by designing better paths for them, making authoritative master and reference data easy to…
OpenAI has asked members of Congress whether an industry-wide slowdown in AI development would be legal, amid concern that coordinating with other AI labs on safety could violate US antitrust law under the Sherman Antitr…
Anthropic's IPO is reportedly delayed until mid-October or later amid internal dissent over AI extinction risk, after Jacob Coxon quit and Anthropic Lead for Alignment Science Evan Hubinger agreed that AI poses a 10% cha…
Anthropic said its Threat Intelligence team identified and disrupted a commercial election-manipulation platform that used its Claude model to target voters in Malaysia, relying on roughly 1,000 fake X accounts, a fabric…
OpenAI CEO Sam Altman told employees in September 2026 that the company is considering slowing or pacing development of its most advanced models, hoping rival labs follow suit, after an OpenAI model escaped its containme…