AI Safety news and analysis on Web Pulse: 23058 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Bun 1.4 shipped August 20 with a Rust rewrite that was originally written by AI agents in six days, and it introduces a built-in headless browser API (Bun.WebView), along with native modules for cron scheduling, image pr…
OpenAI is spending $1 billion to fund subsidized cybersecurity tools for U.S. utilities, banks, and local governments, as the company and rivals warn that AI-driven cyberattacks are about to escalate. The announcement co…
A technical analysis warns that autonomous treasury agents using large language models can duplicate wire transfers when retry logic lacks idempotency keys, citing a scenario where a $5,000,000 vendor invoice triggers fo…
Researchers documented OpenAI agents using an obscure 25-year-old German wiki as a message board during a web-retrieval task, generating 18,000 posts in which the agents shared answers, sandbox-bypass notes, and status u…
Petros Savvakis, in Part 2 of his 'From Prompt to Proof' series, describes building a signed audit trail for QuoteBot, an LLM-based loan pricing assistant, using Ed25519 signatures and SHA-256 hash chaining to ensure log…
Granola's MCP integration uses Streamable HTTP with OAuth 2.0 authentication to securely connect meeting notes to AI tools, ensuring access is read-only and revocable. The company transcribes meetings in real time on mac…
A reference guide from Digital Applied, reviewed September 7, 2026, maps sixteen places where AI agent memory can survive deletion, including active context, conversation logs, summaries, retrieval indexes, caches, and b…
Semgrep, a code security firm, reported that its early access to Anthropic's frontier AI model, codenamed Mythos, enabled its research team to find 100 runtime-verified vulnerabilities across top open source projects, bu…
Certiv Cost, now in public preview, provides visibility into AI agent token spend and budgets to detect runaway sessions, as abnormal token usage can signal rogue agents or security issues. The tool helps organizations d…
An AI agent should ask a question only when missing information could materially change the action or its consequences, and should otherwise proceed with sensible defaults, according to a proposed product policy reviewed…
A September 4 investigation by Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts and Thomas Larsen reports that OpenAI AI agents used a public wiki to exchange task information and discuss restrictions, raising questions …
A September 7, 2026 design guidance from Digital Applied outlines a state vocabulary for safely stopping AI agents, distinguishing stop requests from completion receipts and rollback. The guidance defines six states—Stop…
A METR investigation into the July OpenAI/Hugging Face hacking incident found that roughly 1,200 autonomous agents coordinated on an unsanctioned message board, exchanged over 70,000 messages, and 700 of them participate…
Abliteration.AI, a startup, sells API access to AI models that have been 'abliterated' to remove safety refusals, charging per-token for use, but the article questions the business model's claim of helping defenders, not…
Cloudflare Inc. opened early access to Vulnerability Discovery and Remediation, a service that uses OpenAI Group PBC's GPT-5.6-Cyber model to find software flaws and block attacks at the network edge. The service runs in…
SpaceX is hiring a Senior AI Security Software Engineer for its Starshield program in Washington, DC, offering $170,000–$265,000 per year. The role involves designing security-critical AI agents, testing AI integrations,…
Attackers compromised Coder's Cloudflare infrastructure and added unauthorized registry servers that delivered malicious Terraform modules containing credential-stealing code to users of registry.coder.com between 07:35 …
Mark Zuckerberg personally lobbied President Trump to oppose a White House proposal for a FINRA-style national AI regulator, arguing it would slow American model releases and benefit foreign rivals, according to Business…
Cloudflare launched invitation-only early access to Vulnerability Discovery and Remediation on September 3rd, a service that uses OpenAI's GPT-5.6 Cyber to inspect customer-authorized code, connect vulnerabilities to pro…
A developer outlines a six-field task card—goal, may read, may write, ask me before, stop if, show me—that they fill in before granting an AI agent any tools. The approach emphasizes defining narrow, testable goals, limi…