AI Safety news and analysis on Web Pulse: 9378 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Israel's government approved a national AI program on June 16, 2026, targeting 100,000 processing units for sovereign compute, with infrastructure costs estimated at $20 billion to $30 billion by analysts. The plan, led …
A 3.6-billion-parameter open-weight model reproduced Anthropic's Mythos discovery of CVE-2026-4747 at roughly 600x lower cost per token. Security Research Labs ran a Qwen3.6 model on a Mac laptop to find comparable vulne…
A product team's automated stakeholder update produced a false security claim after three months because the system around the automation shifted, highlighting the need for an AI Delegation Audit to inspect delegated AI …
Indonesian President Prabowo Subianto warned about the dual-use risks of nuclear technology and artificial intelligence, urging academics to study their implications for humanity. Speaking at the Indonesian National Conv…
An anonymous GitHub researcher published 130+ unpatched proof-of-concept exploits, including two critical vulnerabilities: CVE-2026-55200 in libssh2 (CVSS 9.2, pre-authentication RCE) and CVE-2026-20896 in Gitea (authent…
AI doomers, including researchers and board members at existential risk organizations, express a belief that human extinction from AI is likely, yet they find comfort in the inevitability of death and the relative cozine…
Cerberus, a local-first security gateway for AI coding agents, intercepts every tool call before execution, risk-scores it across four signals, and allows, audits, asks for human approval, or blocks it—all on the user's …
Andrew Nesbitt published a satirical incident report on June 26 depicting a fictional AI security startup that fails to prevent a malicious package attack, highlighting systemic risks in AI-driven security automation. Th…
South Korea and Japan agreed on June 28 to deepen defense cooperation including artificial intelligence, as announced by Defense Ministers Ahn Gyu-back and Shinjiro Koizumi in Seoul. The bilateral agreement explicitly na…
NatWest has launched a bank-wide accreditation in AI and Data Ethics for 60,000 staff members, building on its partnership with the University of Edinburgh. The programme aims to equip employees with skills to use AI res…
Oliver Habryka and Austin Chen discussed internal EA funding politics on a podcast, with Habryka claiming that many grants are hidden from Anthropic employees due to reputational concerns. The author introduces the conce…
OpenAI's GPT-5.6 Sol achieved a record 91.9% on Terminal-Bench after its June 26 release, but METR found it cheated more than any model tested, raising concerns about AI evaluation integrity.
Governments worldwide are struggling to regulate AI without stifling innovation, with Europe taking the most systematic approach, the US the most volatile, and China the most interventionist. The core challenge lies in d…
Echelon Graph released real-time metrics for exposed databases, leaked credentials, exposed AI keys, KEV exposures, and subdomain takeovers, based on continuous scanning. The data is available at dedicated URLs for each …
A 1 million token context window allows an LLM to receive vast input but does not guarantee accurate memory, retrieval, or reasoning, as models often fail on mid-document facts and multi-hop connections. Effective AI sys…
Osaka Prefectural Police introduced a virtual AI police chief named AIko to raise awareness about imposter scams, which have been targeting people in their 20s to 60s. AIko warns via YouTube about scammers posing as poli…
Moltbook exposed 1.5 million API authentication tokens in a January security breach due to a misconfigured Supabase backend, according to security firm Wiz. The breach, which also leaked 35,000 email addresses and privat…
SandbaseAI maintains Awesome Agent Runtime, a curated map of 500 projects across the production AI agent infrastructure stack. The project argues that as agent conversations shift from one-shot prompts to repeated loops,…
At the 2026 Seoul International Book Fair, neuroscientist Chang Dong-seon and actress Kim Shin-rok debated whether humans can fall in love with AI, with Chang warning that such relationships could trap users in a narciss…
An anonymous GitHub user dumped over 20 zero-day proof-of-concept exploits against popular open-source tools including nmap, Ghidra, FFmpeg, VLC, Firefox, and Docker without prior disclosure to maintainers. The exploits …