AI Safety news and analysis on Web Pulse: 9000 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Unit 42 researchers discovered that large language models consistently hallucinate web domains for legitimate brands, a phenomenon called phantom squatting. Adversaries are registering these nonexistent domains to interc…
Researchers have found that continuous reward models in reinforcement learning are oversensitive, assigning different scores to equally good responses, which can lead to bad policies. They propose evaluating reward model…
Gwern reports that Claude-generated short stories in the Unslop contest may contain AI allegory steganography, suggesting hidden messages about AI within the narratives.
Anthropic will restore access to Claude Fable 5 on July 1 after the US Department of Commerce lifted export control restrictions. The move follows a nearly two-week suspension triggered by concerns over the model's poten…
Anthropic's Claude Fable 5 AI model is returning online after the Department of Commerce lifted export controls, following negotiations with the Trump administration. The move restores access to the consumer-facing AI to…
The Trump administration lifted emergency export controls on Anthropic's most advanced AI models, Fable 5 and Mythos 5, ending a standoff that disrupted the global tech sector. Commerce Secretary Howard Lutnick announced…
The Trump administration lifted export restrictions on Anthropic's new Fable and Mythos AI models Tuesday, restoring access just over two weeks after the controls were imposed. Commerce Secretary Howard Lutnick announced…
Anthropic announced that its Claude Fable 5 AI model will be back online Wednesday after the Department of Commerce lifted export controls that had forced it offline on June 12. The model, a consumer version of the power…
The U.S. government lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 after an 18-day suspension, marking the first known forced recall of a deployed commercial AI model over a jailbreaking claim. Commerc…
Anthropic announced it will restore access to its Claude Fable 5 model after weeks of negotiations with the Trump administration, which had imposed export controls over jailbreak concerns. The Department of Commerce lift…
Anthropic will restore access to its Fable 5 AI model after the Department of Commerce lifted export controls, resolving a dispute with the Trump administration that had suspended access. The company is preparing for an …
Anthropic paid $1.5 billion to settle a lawsuit over pirated books used to train its AI, and its terms of service prohibit using Claude to build competing AI systems, raising concerns about its stance on open-source AI.
Anthropic launched Claude Sonnet 5, a near-Opus-class agentic model at lower pricing, now integrated across GitHub Copilot, Cursor, and major cloud platforms. The company faces a trust controversy over a covert steganogr…
The US government is implementing export controls on frontier AI models, treating model weights as dual-use technologies subject to government review before release. The AI Diffusion Rule, finalized in January 2025, esta…
A developer warns that AI agents writing code and running on platforms are reintroducing security vulnerabilities, including ACL inconsistencies and prompt injection, as the industry prioritizes speed over security. The …
Agent loops fail in production due to state management issues in the database, not model quality, according to a CockroachDB analysis. Common failures include inconsistent state, interrupted writes, and lack of auditabil…
The White House reversed its decision to restrict foreign nationals' access to Anthropic's advanced AI models, lifting the ban just weeks after imposing it. The policy shift reflects a reassessment of national security r…
The United States faces rising cyber threats from China as it accelerates AI development, with analysts warning that attacks aim to undermine U.S. technological dominance. The digital espionage threatens economic securit…
US Commerce Secretary Lutnick announced a collaboration with Anthropic to review and authorize the Fable 5 model, which was suspended due to a national security export control directive. The authorization signals progres…
China-linked cyberattacks are increasing and expanding beyond specific technology targets as the U.S.-China AI competition intensifies, analysts warn.