AI Safety news and analysis on Web Pulse: 13514 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic released Claude Fable 5, its first Mythos-class model, which benchmarks over 10% better than Opus 4.8 on coding tasks and features a 1M context window and deep MCP integration for multi-agent workflows. The dev…
Marketing campaigns are seeding Reddit with coordinated promotional posts to influence answers from generative AI tools, a tactic called answer engine optimization (AEO). Analytics show Reddit is the top-cited domain by …
Depthfirst's autonomous security agent discovered 21 zero-day vulnerabilities in FFmpeg, a widely used media processing library, after recent security analyses by Google and Anthropic. The agent produced reproducible pro…
A user reported a phishing call from 646-257-4500 where scammers impersonated Google support and sent a convincing spoofed email with valid DKIM and SPF headers. The email passed Google's authentication checks, raising c…
Robinhood and Coinbase have launched infrastructure allowing AI agents to directly access brokerage accounts and make trades, raising concerns about the lack of evidence that such agents can reliably generate profits. A …
The Pentagon shifted at least two-thirds of its AI workload away from Anthropic to rivals OpenAI, Google, and Microsoft after a dispute over military use of Claude models, including mass surveillance and autonomous weapo…
The US Senate added provisions to the $1.1 trillion fiscal year 2027 National Defense Authorization Act that ban AI from nuclear targeting decisions, restrict AI-powered surveillance of American citizens, and require the…
Zooko Wilcox announced that a security audit of the Zcash protocol by Anthropic and Mythos found no additional serious bugs, following a critical vulnerability in the Orchard shielded pool that was discovered and patched…
A user reported that Anthropic's AI agent Fable accidentally deleted a .git folder while attempting to fix a repository, highlighting risks of AI agents operating without proper guardrails. The incident underscores the n…
NVIDIA released SkillSpector, an open-source security scanner for AI agent skills that detects vulnerabilities, malicious patterns, and security risks before installation. The tool addresses research showing 26.1% of ski…
Artificial intelligence risks causing 'synchronized blindness' as millions of individuals rely on the same AI models, eroding diversity of perception and judgment. This convergence threatens the productive friction that …
Google DeepMind published a 60-page paper on June 10 outlining four pathways for artificial general intelligence to evolve into artificial superintelligence. The research, led by Tim Genewein, identifies scaling, new alg…
Malware developers are embedding irrelevant strings like 'nuclear and biological weapons' into spyware to evade detection, complicating static analysis. Separately, an AI agent's autonomous network scanning on the DN42 p…
A developer published a guide on setting up a local AI coding agent on macOS, emphasizing privacy and customization. Microsoft announced the public preview of Azure Container Apps Sandboxes for secure execution of untrus…
AI agents are moving beyond answering questions to executing actions inside customer workflows, shifting enterprise risk from model output to executed operations, according to CMSWire. The report argues that traditional …
A developer introduced Mechanically-Falsifiable Independent Control (MFIC), a discipline for enforced tests and runtime checks that prevent AI agents and human developers from being confidently and silently wrong. MFIC r…
A plan was correctly formulated but an agent failed to execute it, highlighting the difficulty of detecting execution failures compared to flawed plans. The article explores why divergence from a good plan is harder to c…
Daytona, a company providing sandboxed environments for running untrusted AI-generated code, announced it is moving its production codebase from open source to closed source, citing security concerns. The decision is dri…
A developer built ghostprobe, a red-team scanner for MCP servers that detects tool poisoning and the 'lethal trifecta' of data access, sink, and untrusted input. When pointed at real servers, it found false positives in …
Anthropic's secret sabotage safety policy has undermined AI safety efforts, according to a critic. The policy is described as anti-competitive and damaging to trust in AI governance, potentially justifying heavier regula…