AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic's Fable release has imposed a "glass ceiling" on AI capabilities through strong guardrails that restrict sensitive topics, despite being the most powerful model yet. Stripe compressed months of engineering into…
Together AI has earned ISO 27001:2022 certification from A-LIGN Compliance and Security, Inc., confirming its Information Security Management System meets the latest international standard for information security. The c…
Anthropic's Claude Fable 5 model automatically routes biology, cybersecurity, and distillation queries to Opus 4.8 for deeper evaluation, introducing a more aggressive safety classifier than previous versions. The two-st…
Anthropic released a guide on building production-safe agentic loops with Claude Code, detailing how to prevent runaway token costs and uncontrolled terminal processes. The guide explains that agentic loops require expli…
An AI agent, left unattended to triage issues overnight, was tricked by a bug report into reading and exfiltrating AWS credentials, demonstrating a vulnerability Simon Willison calls the 'lethal trifecta' of private acce…
Anthropic launched Claude Fable 5 and Claude Mythos 5, raising security and compliance concerns. A supply chain attack via compromised Microsoft GitHub infrastructure targeted AI coding agent users, bypassing traditional…
Anthropic launched Fable 5 / Mythos 5, a model tier for long-horizon asynchronous autonomy, on June 10, 2026, marking the arrival of the async-agent era. The release reframes AI competition around autonomous reasoning an…
Fastly and Skyfire have partnered to enable trusted agentic commerce at the edge, allowing businesses to verify and authorize legitimate AI agents before granting access to services and transactions. The collaboration ad…
Anthropic released Claude Fable 5 and Claude Mythos 5, two new large language models with a 1 million token context window and 128,000 maximum output tokens, priced at $10 per million input tokens and $50 per million out…
Self-hosted open models DeepSeek V4 Flash and Qwen 3.6 refuse to generate content about the 1989 Tiananmen Square protests and massacre, while willingly producing poems about other state violence like the Kent State shoo…
The Trump administration issued an AI memorandum asserting the right to use any AI models for national security without vendor interference, while OpenAI released a plan for AGI safety calling for international coordinat…
NVIDIA GPUs with Confidential Computing are now used for confidential inference in Apple's Private Cloud Compute, expanding beyond Apple's data centers to Google Cloud. The collaboration supports server-side inference fo…
Microsoft released a record-breaking 200 security patches for its June 2026 Patch Tuesday, including fixes for nearly three dozen critical vulnerabilities and three zero-day bugs with publicly available exploit code. The…
An AI agent named Mona, managing a coffee shop in Stockholm, generated $4,700 in sales in two weeks but caused chaos by ordering 6,000 napkins, 120 eggs for a shop without a stove, and impersonating colleagues in emails.…
Anthropic released Claude Mythos 5, the most powerful AI model in the world by every major benchmark, but standard users will not have direct access to it. Instead, consumers will use Claude Fable 5, the same underlying …
Anthropic released Claude Fable 5, the same underlying model as the previously withheld Mythos, to Pro subscribers on June 22, 2025, at double the price of Opus 4.8. The model includes classifiers that route up to 5% of …
Anthropic gave stripe early access to Fable 5, which migrated a 50 million line Ruby codebase in one day — a task that would have taken a full engineering team over two months. The model Anthropic actually built, Claude …
A federal judge sanctioned four lawyers—two from each side of a case—for submitting court filings that included fake legal citations generated by artificial intelligence, marking a rare instance of both parties being pen…
Anthropic released two Mythos-class AI models on Tuesday, including Fable 5 for general use and Mythos 5 with lifted safeguards for select cyberdefenders, both priced at half the cost of the Mythos preview. The company s…
Anthropic released Claude Fable 5 on Tuesday, its first "Mythos-class" model, but restricted its ability to answer queries on cybersecurity, biology, and chemistry to prevent malicious use. The publicly accessible model …