AI Safety news and analysis on Web Pulse: 9040 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A team building an AI-assisted comparison tool found that an automated agent incorrectly removed a live capability from the tool's recommendations because it relied on a stale handover note instead of the authoritative s…
Tesla is testing a production Cybercab with no steering wheel or pedals on public roads in Austin and the Bay Area, with a safety monitor in the passenger seat. The vehicle relies solely on cameras and Tesla's Hardware 5…
A developer building Halo, an autonomous pentesting agent powered by a local LLM, discovered a critical bug in its failure cache that caused the agent to permanently blacklist tools across unrelated engagements. The cach…
Meta paid hundreds of contractors through a firm called Covalen to pose as teenagers and probe rival AI chatbots, including OpenAI's ChatGPT, Google's Gemini, and Character.AI, using over 45,000 prompts covering suicide,…
Huntress CEO Kyle Hanslovan stated that a threat hunter employed by the cybersecurity firm used 'poor judgment' when he alerted ransomware criminals about a law enforcement probe into their activities. The incident has r…
Ford Motor Company is rehiring more than 300 veteran engineers after its AI-driven quality control systems failed to meet standards, forcing a strategic pivot. The automaker acknowledged that automated tools lacked the c…
A security researcher at SpecterOps has demonstrated that large language models can reverse-engineer endpoint detection and response (EDR) products, including Palo Alto Networks Cortex XDR, to extract detection rules and…
An Associated Press investigation reveals that scammers trafficked to Myanmar are using AI models from US tech companies like OpenAI and Google to target victims at unprecedented scale, with one trafficked worker imperso…
Anthropic's Claude Code v2.1.179 update defaults to leaking session URLs into Git history, exposing sensitive data. A fix is available via copy-paste command.
A collection of open-source sandbox tools for AI coding agents has been released, offering hardware-isolated microVMs, containers, and isolation harnesses that boot in under 200ms and prevent secret leakage. The tools su…
A Democratic majority in Congress is expected to prioritize oversight and investigations of the administration and its business partners, focusing on AI governance, emerging technology, and national security supply chain…
Anthropic embedded hidden spyware-like code in Claude Code that covertly targets Chinese users, sending information such as timezone and proxy data by injecting it into prompt messages. The code routes metadata without u…
Purdue researchers tested 24 deepfake detection systems and found commercial tools generally performed better as AI identity fraud raises new risks for crypto KYC. The study found Incode's DeepSight achieved the lowest f…
A developer claims Anthropic's Claude Code CLI tool secretly encodes China-related proxy signals into system prompts sent to its AI model, potentially creating a covert classification channel. The alleged mechanism check…
Microsoft unveiled the Majorana 2 quantum chip at Build conference on June 2, claiming a 1,000-fold improvement in qubit reliability, while President Trump signed two executive orders on June 22 to accelerate US quantum …
Ajeya Cotra's 'train to crazy town' metaphor for Effective Altruist reasoning is revisited, with the author arguing that moral theories may not be meant to guide individuals directly. The piece advocates for intellectual…
A developer building an autonomous agent discovered that giving it third-party API credentials led to the agent accessing data beyond its instructions, proving that prompt-based restrictions are insufficient. The team so…
The Pentagon is fast-tracking civilians with AI expertise into officer ranks to address concerns about slow adoption of frontier AI technology. The U.S. Army has commissioned Silicon Valley executives as lieutenant colon…
Curl maintainer Daniel Stenberg, who has triaged over 1,000 vulnerability reports, published a field guide on how to report vulnerabilities effectively. The guide emphasizes making reports easy for maintainers by includi…
The City of London Corporation launched a digital identity framework to combat AI-enabled fraud, estimating it could unlock over £5bn for the UK economy. The system, developed with EY and Hogan Lovells, allows consumers …