AI Safety news and analysis on Web Pulse: 13161 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Sync, a new project-management and quality-control system for AI coding agents, launched in alpha v0.4.5. The tool runs locally and enforces agent accountability to defined specs, rules, and roadmaps, flagging stale cont…
The US proposed the Baruch Plan in 1946 to place atomic energy under international control, but the USSR vetoed it, fearing loss of nuclear development. The plan's failure led to a nuclear arms race, highlighting challen…
Nylas has introduced spam detection policies for its Agent Accounts, which are mailboxes built for AI agents and system identities. The policies filter spam at the mailbox layer before the agent processes messages, using…
Nylas introduced Agent Accounts, a hosted mailbox with six system folders that provides an immutable audit trail for email agents. The sent folder stores every outbound message as a real object, timestamped and fetchable…
A team's support triage agent suffered a prompt regression that caused it to reply to all emails for hours, highlighting the need for least-privilege access control in AI agent fleets. The Nylas security guide recommends…
A new study from FusionAuth reveals that 65% of organizations suffered an AI identity breach in the past year, with those most confident in their security being breached at the highest rate (84%). The report warns that c…
A machine-learning screen of 2.6 million cancer papers found nearly 10% show signs of paper mill fraud, while fabricated citations in biomedical papers have grown twelve-fold in two years, reaching one in 277 papers in e…
The U.S. government issued an export control directive requiring Anthropic to suspend access to its top-tier AI models, Fable 5 and Mythos 5, for foreign nationals over national-security concerns. In response, a Times of…
UC Berkeley professor Hany Farid, a leading digital forensics expert, told the New York Times he struggles to distinguish AI-generated media from real content, calling it 'like going blind.' Farid is leaving Berkeley to …
GitHub now allows pull requests created by github-actions[bot] to trigger CI/CD workflows, but only after a maintainer with write access explicitly approves the PR. The change, effective June 11, fixes a bug that previou…
A privacy issue in Mozilla Thunderbird leaks users' language preferences in MIME message headers, exposing enabled spellcheck languages via the Content-Language header. The issue persists even when spellcheck is disabled…
Enterprise teams replicating the software development lifecycle with AI agents are importing defenses against human flaws that models don't have while missing model-specific failure modes like hallucination and sycophanc…
Hosted coding agents are shifting observability from an enterprise add-on to a core product feature. AWS's Bedrock AgentCore exemplifies this trend by enabling agents to run remotely, which requires replacing developer i…
Government-imposed AI model bans are becoming more common, with countries like Italy, Australia, and South Korea restricting access to models such as ChatGPT and DeepSeek. To avoid workflow disruptions, organizations mus…
The Forum of Incident Response and Security Teams (FIRST) released its 2026 Mid-Year Vulnerability Forecast, revealing actual CVE disclosures are running 46.3% above projections from four months ago, with a revised full-…
LLM guardrails control risk by constraining model behavior through input validation, output filtering, and fact-checking. Key strategies include prompt sanitization, length limits, content filtering, response validation,…
1Password acquired Apono to extend its identity security platform with just-in-time privileged access governance, addressing the growing challenge of managing access for human, machine, and AI identities. The acquisition…
The US government issued an export-control order disabling Anthropic's Fable 5 and Mythos 5 models after a jailbreak, while New York's attorney general subpoenaed OpenAI over advertising and data practices. These actions…
OpenRouter launched Fusion API, a compound AI system that orchestrates up to eight LLMs in parallel to improve reasoning accuracy, achieving 69% on the DRACO benchmark. Microsoft released seven new MAI models built from …
A new paper titled 'When AI Says It Feels' warns against mistaking eloquent AI language for evidence of consciousness, urging a multi-angle approach to evaluating machine sentience. Researchers argue that claims of machi…