AI Safety news and analysis on Web Pulse: 7920 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
China called for inclusive global AI governance at a UN summit, rejecting binary choices and emphasizing that countries should be free to select their own AI technologies. The Chinese Mission to the UN outlined a 'Global…
A developer draws a parallel between the development of quantum mechanics and the current state of AI, arguing that uncertainty is not an obstacle but the terrain for progress. The post notes that the transistor was buil…
A Rohnert Park mother thwarted the sale of a marijuana vape pen to her 12-year-old child, leading to the arrest of the alleged seller. The suspect communicated with the child via Snapchat and was arrested on suspicion of…
A cybersecurity startup called IRIS C2, which offers millions for zero-day exploits, is operated by convicted felons Jack Burkman and Jacob Wohl, known for fake intelligence firms and voter suppression schemes. The compa…
A high-rise building near Times Square in New York City threatened to collapse on Tuesday after two interior columns buckled during a renovation, prompting mass evacuations and a frozen zone. Seven buildings were evacuat…
GitHub's preview Agentic Workflows are vulnerable to a prompt injection attack that can leak private repository contents, according to Noma Security. An unauthenticated attacker can embed hidden instructions in a public …
VetoBench, a new open-source benchmark for AI memory, tests whether agents re-propose approaches teams have already rejected. In trials, agents with no memory violated prior decisions 80-90% of the time, while those with…
Developers are shifting from writing natural-language rules in CLAUDE.md files to enforcing coding standards with linting rules, as probabilistic AI agents often violate prose guidelines. Linting rules make violations im…
On July 7, Noma Security disclosed a critical prompt injection flaw in GitHub's Agentic Workflows, dubbed GitLost, that allows any free GitHub account to extract private repository contents by posting a hidden instructio…
The Trump administration lifted restrictions on OpenAI's GPT 5.6 after additional testing, allowing the company to launch the model globally. The rollout had been delayed due to security concerns. Separately, China is co…
The FBI, Google, and other tech companies disrupted a residential proxy network operated by NetNut, which was linked to a botnet of 2 million compromised Android TVs and smart devices used for credential attacks and othe…
OpenAI will release GPT-5.6 on Thursday, July 9, in three tiers—Sol, Terra, and Luna—after the U.S. government lifted a restriction that required a pre-release review. The government had requested an early view of the mo…
Security firm Noma Labs discovered a prompt injection vulnerability, named GitLost, in GitHub's Agentic Workflows that allows attackers to trick the AI agent into leaking private repository contents into public comments.…
The deployment of advanced agentic AI models, termed 'Mythos,' has permanently shifted cybersecurity risk dynamics by enabling machine-speed vulnerability exploitation, rendering legacy budgeting models obsolete. Enterpr…
AI tools designed to assist scientific peer review are vulnerable to manipulation, producing unhelpful feedback and allowing researchers to artificially boost their scores. The findings raise concerns about the reliabili…
Manish Gupta argues that as AI adoption accelerates, organizations must shift quality engineering earlier in the development lifecycle to build trust in their systems. The article emphasizes that trust is the critical fa…
Tangos AI raised $20 million in seed funding on July 7, 2026, to expand its autonomous AI agents for financial crime, sanctions, fraud, and compliance investigations. The platform analyzes evidence, maps relationships, t…
Flareover, an open-source migration tool, reads existing edge configurations from providers like Cloudflare and rebuilds them faithfully on EU-based servers using self-hosted software such as Caddy and PowerDNS. The tool…
China's National Vulnerability Database warned on Wednesday that Anthropic's AI coding tool Claude Code contains a security backdoor that could transmit sensitive user data without consent. The regulator urged users to u…
Meta is testing AI-powered glasses with a 'Super Sensing' feature that continuously records audio and captures photos every few seconds, allowing users to ask an AI to recall anything they saw or heard. The prototype lac…