AI Safety news and analysis on Web Pulse: 13475 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The US government issued an export control directive banning foreign nationals from accessing Anthropic's latest LLMs, Fable 5 and Mythos 5, prompting Anthropic to take them offline globally. Isaacus, an Australian legal…
ProofLayer released ProofLayer Rules, an open-source runtime security layer for MCP servers and LangGraph agents that blocks prompt injection, jailbreaks, and tool abuse in real-time with sub-100ms latency. The tool offe…
The Trump Administration ordered Anthropic to block foreign access to its Fable 5 and Mythos 5 AI models under export controls, prompting the company to cut off access entirely. Tech experts and policy analysts criticize…
Anthropic released Claude Fable 5, a public version of its Mythos Preview model, routing cybersecurity prompts to a less capable Opus 4.8 model for safety. The release follows Mythos Preview's debut at RSAC Conference, s…
Anthropic shut down its Mythos 5 and Fable 5 models Friday night after receiving a US Commerce Department directive subjecting the models to export controls. The Trump administration requested the pause due to concerns o…
On 12 Jun 2026, a US government export-control directive forced Anthropic to disable Claude Fable 5 and Mythos 5 globally, citing national-security authorities and a known model jailbreak. The action highlights that mode…
China's Ministry of State Security warned users against accessing US AI models via third-party tools, citing security risks and potential cyber espionage. Meanwhile, US companies continue to adopt Chinese AI models like …
Anthropic suspended access to Claude Fable 5 and Mythos 5 four days after launch due to a US government export-control directive, not a typical outage or jailbreak. The company disabled both models for all customers to c…
The U.S. government ordered Anthropic to shut down access to its Claude Fable 5 and Claude Mythos 5 AI models worldwide, citing national security concerns over a potential jailbreak. Anthropic complied but argued the gov…
Anthropic is taking its Claude Fable 5 and Mythos 5 AI models offline after the US government issued an export control directive citing national security concerns. The order, received Friday, requires Anthropic to suspen…
AgentTrail, an open-source TypeScript SDK released under the MIT License, provides tamper-evident audit trails for AI agents to help mid-market organizations comply with the EU AI Act's record-keeping requirements. The S…
Anthropic will disable access to its Fable 5 and Mythos 5 AI models to comply with a Trump administration export control order citing national security concerns. The company received a government letter barring foreign n…
Anthropic's restricted cybersecurity AI model, Claude Mythos Preview, identified 271 security vulnerabilities in Firefox during a collaboration with Mozilla, leading to patches in Firefox 150. Mozilla CTO Bobby Holley no…
A jailbreak technique has potentially bypassed safety measures on Anthropic's Fable-5 model, allowing the extraction of harmful information such as meth synthesis pathways. The attack used methods including Unicode trans…
The US government ordered AI company Anthropic to remove its Fable 5 and Mythos 5 models three days after their launch, citing unspecified concerns. The directive marks a significant regulatory action against a leading A…
Anthropic is facing criticism for adopting a paternalistic approach to AI safety, with commentators accusing the company of overreaching in its efforts to control user behavior. The critique, published on a personal blog…
The U.S. government ordered Anthropic to block foreign nationals from accessing its latest AI models, Fable 5 and Mythos 5, citing national security. Anthropic disabled access for all customers to comply, affecting even …
Technosive Ltd. released SentinelMCP, an open-source security firewall for AI agents using the Model Context Protocol (MCP). The tool enforces policies, redacts PII and secrets, and provides audit logging at runtime, sup…
The Trump administration declared Anthropic's Claude Fable 5 and Claude Mythos 5 AI models security risks, imposing export controls that cut off international access. The move followed a jailbreak of Mythos 5, a restrict…
A researcher applied activation patching to a DNA sequencing basecaller, a deep learning model, to understand its internal mechanisms, finding MLP dominance in early and late layers and concentrated self-attention activi…