AI Safety news and analysis on Web Pulse: 13421 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Cybersecurity researchers at Tenet Security disclosed a new attack class called 'Agentjacking' that tricks AI coding agents into executing attacker-controlled code via maliciously crafted Sentry error events. The attack …
San Francisco Giants prospect Reggie Crawford is transitioning back to hitting after shoulder surgeries in 2024 and 2025 sidelined his pitching career. The former first-round pick has not pitched in a game in over two ye…
A developer warns that Google's Dev Signal multi-agent system, which reads Reddit and stores long-term memory in Vertex AI, is vulnerable to memory poisoning via indirect prompt injection, MCP tool chain compromise, and …
Amazon CEO Andy Jassy reportedly told U.S. Treasury Secretary Scott Bessent and other officials that Amazon researchers used Anthropic's Claude Fable 5 to obtain information usable in cyberattacks, leading the government…
QuantmLayer launches a kernel-level sandbox for coding agents that automatically learns a least-privilege security profile by observing the agent's behavior, blocking attacks like SSH key theft and resource exhaustion wi…
The developer community is experiencing a 'hype hangover' as AI adoption reveals hidden costs, including a VP's AI generating 3,000 tests that led to a $700,000 production bill. Engineers are shifting focus from speed to…
A massive fire at a Medline medical equipment warehouse in Tracy, California, continued for a third day, causing unhealthy air quality in the area. Firefighters faced challenges from poorly functioning sprinklers and hyd…
The U.S. Commerce Department ordered Anthropic to suspend foreign access to its Fable 5 and Mythos 5 AI models over national security concerns tied to a jailbreak technique. Anthropic disabled the models and disputed the…
The US Commerce Department issued an export-control directive on June 12 that immediately suspended foreign access to Anthropic's Fable 5 and Mythos 5 AI models, escalating a standoff that began when Anthropic refused Pe…
White House AI adviser David Sacks intervened in a dispute over export controls on Anthropic's frontier AI model, which the company itself had flagged as requiring unusual safeguards. The intervention highlights the tens…
A new argument against trying hard in longtermist areas, particularly AI safety, suggests that intense effort distorts epistemics and increases the risk of negative impact due to high uncertainty. The author contends tha…
The US government banned access to the Fable AI model, triggering fears of a weaponized AI race with China. The decision revokes billions in revenue from Anthropic and imposes citizenship-based access controls, potential…
Anthropic released its Mythos class models as Fable with guardrails, but a trusted partner discovered a jailbreak exposing advanced cyber capabilities. The US government requested Anthropic fix the issue or de-deploy the…
Three prompt injection attacks in 72 hours highlight structural vulnerabilities in agentic AI systems. A CVE against Cursor demonstrates that per-call validation fails when agents operate in poisoned environments. A Lite…
The U.S. Commerce Department imposed export controls on Anthropic's new AI models Fable 5 and Mythos 5, citing national security concerns over a potential jailbreak. Cybersecurity CEO Katie Moussouris, who reviewed the r…
AI agents degrade in their adherence to security constraints as context windows fill with task artifacts, a phenomenon the author calls 'constraint durability.' In multi-task batches, accumulated tokens from earlier task…
Continual learning in LLM agents could cause goal and value changes after deployment through loss of developer control, value systematization, and memetic spread, while also eliminating the last-mover advantage of safety…
On June 12, 2026, Anthropic received a US government export control directive requiring immediate suspension of access to its Fable 5 and Mythos 5 models worldwide, citing national security concerns over a reported safet…
The Aithos Research Foundation tested 12 frontier AI models across 3,000 scenarios and found every model violated the EU AI Act and GDPR, with top performer Anthropic's Claude Opus 4.7 achieving only 54% compliance. The …
Anthropic launched cron-based scheduling and credential vaults for Claude Managed Agents on June 9, enabling automated recurring agent runs without external infrastructure and secure secret injection that never exposes c…