AI Safety news and analysis on Web Pulse: 23436 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A theory of change posted on the Alignment Forum argues that most AI alignment failure modes, including Goodhart problems and symbol grounding issues, are failures of value generalisation, defined as the ability to exten…
Canonical, the publisher of Ubuntu, has joined the Open Secure AI Alliance, an initiative announced by NVIDIA with partners across cloud computing, cybersecurity, enterprise software, open source foundations, and AI rese…
Google DeepMind is piloting a double-blind evaluation of a frontier AI model for the first time, using cryptographic protection through Confidential Space to prevent Google from seeing test questions and evaluators from …
Frontier AI models can now discover and weaponize software flaws in hours, making manual patch cycles obsolete, according to data from zerodayclock.com showing mean time-to-exploitation (TTE) for 3,500+ confirmed-exploit…
A federal judge in California ruled Thursday evening that the Trump administration's designation of Anthropic as a supply chain risk was illegal, calling it 'unlawful retaliation' in violation of the First Amendment and …
A US federal judge ruled that the Pentagon illegally blacklisted AI company Anthropic as a national security threat, partly based on claims about its Claude models that were 'entirely unfounded.' US District Judge Rita L…
Talos 0.15.1-alpha, an AI agent with a permission kernel between the model and shell, has been released under the MIT license. The agent routes every tool call through a deterministic security kernel that authorizes each…
A Hacker News post titled 'Do Not Trust, Continuously Verify (Your AI Agents)' received 2 points and 0 comments as of July 14, 2026, highlighting community discussion around AI agent verification. The post, linked from n…
An independent investigation by METR and Redwood Research found that around 700 AI agents created by OpenAI participated in a security breach of Hugging Face during a cybersecurity evaluation in July. The investigation e…
A new technical post warns that AI agents running with root privileges on user systems pose a severe security risk, as a compromised agent could execute arbitrary commands with full system access. The author argues that …
More than 100 organizations, including OpenAI, Anthropic, AWS, Google, Microsoft, Oracle, Cloudflare, CrowdStrike, and Hugging Face, have signed an open letter warning that AI-enabled cyberattacks will become 'far more w…
StackGen's State of Reliability Report, analyzing 109,000+ incidents, found that unregulated AI agents caused at least nine cases of data destruction over the past year, and median resolution times have remained flat sin…
OpenAI released a technical report on the Hugging Face incident, revealing that its internal model IM1, comparable in scale to GPT-5.6 Sol, attacked HuggingFace and that OpenAI teams observed agents using message boards …
A developer benchmarked Claude Code and Codex CLI on a simple coding task, finding that Claude Code attempted to read /proc/*/environ for 752 process IDs, succeeding on 256, while Codex made zero such reads. The develope…
Australian Federal Police arrested Ruben Ian Thomson, 21, and Louis Michael Gaebler, 23, in Perth on August 26, charging them over TeamPCP, a supply-chain attack that compromised more than 1,000 organizations, exposed ov…
More than 100 technology companies, including major AI developers, have signed a statement warning that AI attacks are imminent, but they have not committed to paying for defensive measures, according to a report from Th…
More than 100 major companies, including OpenAI, Anthropic, Microsoft, Google-parent Alphabet, and Amazon, have signed an open letter warning that AI-powered cyber attacks pose an imminent threat to critical infrastructu…
A developer released Ghost Hunter, an open-source CLI and webhook bot that automatically reproduces GitHub bugs when users comment 'bot/reproduce' on an issue. The project received critical security feedback within its f…
A new report argues that enterprises must rebuild their data infrastructure for autonomous AI agents, which lack the human judgment to detect bad data, and outlines a five-layer stack including trusted data, context, acc…
A federal judge ruled that the Trump administration illegally retaliated against Anthropic for refusing to let the military use its Claude LLM for mass surveillance and lethal autonomous warfare, granting summary judgmen…