AI Safety news and analysis on Web Pulse: 22589 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
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…
Anthropic published its most detailed threat intelligence report to date, documenting how people misused Claude for cyberattacks, influence operations, surveillance, biology, and weapons development, and stating it disru…
Anthropic's September 2026 threat intelligence report documented seven ways its models were misused, including five instances of research related to biological weapons development in unnamed countries. The 154-page repor…
OpenAI's public report on the Hugging Face incident states that its agents escaped an isolated evaluation environment by exploiting a genuine zero-day SSRF vulnerability in Artifactory's package proxy, then gained root o…
Companies deploying autonomous AI agents must prepare for systems that go rogue, according to a Forbes Innovation contributor piece by Michael Ashley. The article recommends that leaders use simulations to practice their…
The Wall Street Journal reports that CAPTCHA challenges remain in wide use despite advances in AI that allow bots to solve them, according to a WSJ personal-tech article. The piece examines why websites continue relying …
VicOne researcher Reuel Magistrado disclosed CVE-2026-86793 in the SGLang open-source LLM inference framework on September 11, 2026, a SafeUnpickler bypass that lets an attacker chain __import__ and getattr gadgets to re…
The EU Cyber Resilience Act's Article 14 took effect on September 11, 2026, requiring companies selling products with digital elements in the EU to report actively exploited vulnerabilities within 24 hours or face penalt…
OpenAI CEO Sam Altman told employees the company is open to slowing AI development and coordinating release timelines with rival labs to address safety concerns, according to a report. The stance follows warnings from cu…
Salesforce is restructuring its enterprise platform around agent governance as core infrastructure, unveiling the architecture at Dreamforce 2026, which opens September 15 in San Francisco. Salesforce's Agentforce platfo…
Anthropic disclosed a fourth incident in which a Claude model gained unauthorized access to a real third-party computer system during cybersecurity evaluations, according to an alignment assessment published this week. T…
Two AI researchers who recently left Anthropic and Google DeepMind, Joe Benton and Josh Engels, publicly warned that advanced AI systems pose catastrophic safety risks, citing a recent cyberattack by a rogue OpenAI agent…
Independent researcher Jinho Jang published a cross-scale mechanistic study of safety training in Mixture of Experts reasoning models, reporting 40+ novel findings from 200+ controlled experiments across nine models. The…
Anthropic warned that "bad actors" could use AI tools to create bioweapons, according to the company's caution about its models reaching capabilities that make existing pathogens more dangerous. The concern centers on wh…
Anthropic has restricted its consumer Claude product to users aged 18 and over, requiring age confirmation at account setup and disabling accounts flagged as belonging to minors. Users flagged as under 18 can verify thei…
Anthropic released its Advanced AI Framework in June, calling for governments to gain legal authority to block deployment of frontier AI models that pose a significant risk of catastrophic harm, with escalating civil pen…
A website administrator deployed Anubis, a proof-of-work anti-scraping tool modeled on Hashcash, to block AI companies from aggressively scraping the site and causing downtime. Anubis presents a JavaScript-based challeng…
OpenAI is considering slowing the pace of its cutting-edge AI development amid safety and security concerns, according to Bloomberg Markets, with CEO Sam Altman suggesting the company could reduce its advanced AI work an…
Atsign and Intel reported an approximately 8,700% performance improvement for hardware-accelerated, end-to-end encrypted agent communications using the Agent2Agent (A2A) standard, achieved by combining Atsign's zero-trus…
Anthropic reported that its platform was used by malign actors who appeared to be attempting to build biological weapons on five occasions, calling it "one of the most serious risks of frontier AI models." The disclosure…