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.
Snyk announces Continuous Offensive Security, a new AI-driven pentesting service that leverages LLMs to find context-dependent vulnerabilities like IDOR and authentication bypasses, which traditional heuristic tools cann…
OpenAI will grant the Korean government, public institutions, and private companies access to its latest cybersecurity AI models through its Daybreak initiative, marking the first such partnership in Asia. OpenAI CSO Jas…
A developer warns that teams building with OpenAI Agents Python should design a control plane before adding more agents, as the hard part of production workflows is not the first demo but defining operational boundaries.…
Researchers at the Karlsruhe Institute of Technology (KIT) in Germany have demonstrated that standard WiFi networks can identify individuals with 99.5% accuracy by analyzing unencrypted beamforming feedback data using ma…
WriterzRoom, a multi-agent content management system for regulated industries, has been developed to ensure every AI decision is auditable, constrained, and recoverable. The system treats governance as a structural prope…
Parts of Pope Leo XIV's encyclical *Magnifica Humanitas* were flagged as AI-generated by automated detection tools, according to analyses cited by The Verge. The Verge reports that the detector Pangram estimated 46 perce…
New Zealand Health Minister Simeon Brown announced the government has begun procurement and testing of an artificial intelligence tool to assist in reading breast screening mammograms, with a planned rollout from early 2…
The Department of Homeland Security, FBI and other agencies are warning about rising anti-technology extremism as concerns over AI-driven job displacement and opposition to data center construction intensify. Intelligenc…
World Labs launched AgentKit on March 17, 2026, a system that uses zero-knowledge proofs to let AI agents prove they are operated by a verified human without revealing personal identity. Coinbase released Agentic Wallets…
Formal verification of complex systems can be simplified by expanding their scope, according to a new analysis drawing on decades of engineering experience. Adding additional layers to a formally verified system paradoxi…
Anthropic co-founder Jack Clark reported experiencing a hallucination in which he perceived a "ghost in the machine" after listening to Pope Francis speak about artificial intelligence. The incident highlights the psycho…
Anthropic co-founder Jack Clark reported experiencing a hallucination in which he perceived a "ghost in the machine" after listening to Pope Francis speak about artificial intelligence. The incident highlights the growin…
A developer refactored core execution modules in `core/tools/buildinpublic.py` and `phases/phase4content.py`, replacing sequential log parsing with an asynchronous event-stream filter to reduce memory overhead during hig…
An analysis by Linch Zhang, posted on the forum LessWrong, found that portions of Pope Leo XIV's encyclical *Magnifica Humanitas* were between 40 percent and 100 percent likely written by AI, according to the AI detector…
Anthropic reported that incorporating a tool allowing its AI model Claude to pause and recall its ethical commitments reduced misaligned behavior on internal evaluations, though researchers cannot determine whether the i…
A Martinez mother lost approximately $5,400 after scammers used artificial intelligence to clone her daughter's voice during a fake kidnapping call in May. Deborah Del Mastro wired the money over five hours after hearing…
An East Bay mother, Deborah Del Mastro of Martinez, was scammed out of approximately $5,400 after callers used AI-generated audio to mimic her 37-year-old daughter's voice and claimed she had been kidnapped. The scammers…
New research shows that post-training alignment makes large language models less human-like in their responses, raising questions about whether this drift is intentional or optimal. A study introducing the "Pinocchio dim…
Preventing malicious AI use faces jurisdictional gaps, as rogue states and lawless regions shield bad actors from existing laws. Open-source AI models complicate monitoring and denial efforts because they can run anywher…
Engineers who passively accept AI-generated code without interrogation are accumulating technical debt that compounds in production failures, according to a new analysis. The solution is not reducing AI use but adopting …