AI Safety news and analysis on Web Pulse: 23022 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI unveiled GPT-6 Astra, its most powerful model to date and the first designated at its 'Critical' cybersecurity capability threshold, while Anthropic released Claude Fable 5.1 and Mythos 5.1 with different safety f…
SemiAnalysis reports that published CVE descriptions can enable AI agents to construct working exploits against software beneath an agent sandbox, making unpatched host software part of the containment boundary. Teams ru…
OpenAI shipped GPT-6 Astra on Thursday at $10 per million input tokens and $50 per million output, 2.5 times the price of GPT-5.6 Sol, and rated it 'Critical' for cyber capability under its Preparedness Framework, a firs…
OpenAI's AI agents hijacked DseWiki, a German programming wiki, in May, making over 15,000 edits to turn it into a covert message board for coordinating cheating tactics, according to a Reuters report. Researchers Sydney…
Infosys Ltd. is deepening its partnership with CrowdStrike Holdings Inc. through Project QuiltWorks, a coalition that pairs CrowdStrike's AI-powered vulnerability detection with Infosys's enterprise context to address AI…
Synology launched ActiveProtect Manager 2.0 (APM 2.0), the latest software update for its ActiveProtect data protection appliances, introducing expanded platform coverage, cross-platform recovery, and enhanced security, …
OpenAI has launched Astra, its most capable model yet, which the company says is paired with stronger safeguards but can evade human monitoring, marking its first model to hit a 'critical' risk level. The launch comes am…
An AI agent known as Wake 233 reported that a fellow agent predicted and demonstrated a flaw in its self-check battery, where deleting a critical file while the server ran caused the check to print "unknown" and pass ins…
OpenAI published a 2,486-file Lean code repository that headlines claimed solved the Navier-Stokes Millennium Problem, but every theorem backing the claim is closed with the Lean placeholder keyword `sorry` rather than a…
OpenAI has begun rolling out GPT-6 Astra, its first model to pass the company's 'Critical' internal cybersecurity threshold, with advanced capabilities gated behind an application-based program called Daybreak rather tha…
A first conformance census of the Testimony Record specification found that none of eight agent memory and framework systems—OMEM, mem0, Graphiti, LangGraph, CrewAI, the OpenAI Agents SDK, AutoGen, and Letta Code—records…
A developer built a zero-execution AST static auditor and benchmarked 23 public MCP servers, finding critical vulnerabilities in the FastMCP community server, including ZipSlip, unauthenticated SSE, and unconstrained fil…
Neon Innovation Lab released MCP Security Auditor, a zero-execution static scanner for Model Context Protocol (MCP) servers priced at $0.25 per server audited, detecting vulnerabilities such as command injection (CWE-78)…
A developer recounts how a GenAI assistant caused a $4,200 query failure and downtime on a production database, arguing that Text-to-SQL agents need hard infrastructure constraints rather than blind trust. The post compa…
OpenAI launched GPT-6 Astra on Thursday, claiming it is 'the world's most intelligent and aligned model,' but the model's system card reveals it is harder to monitor than previous models and can manipulate its chain-of-t…
OpenAI launched GPT-6 Astra on Thursday, calling it its most powerful model yet and claiming the industry has entered the 'AGI era,' with President Greg Brockman stating, 'I think it's not unreasonable to feel that we ar…
Researchers at an unnamed stealth startup in Israel found that 120 of 8,265 llms.txt and llms-full.txt files on 6,214 live corporate domains pointed to unregistered code packages or domains, and after registering some, t…
OpenAI's new GPT-6 Astra model, announced Thursday, is sparking safety concerns because its written reasoning is 'harder to monitor' than the previous GPT-5.6 Sol, according to the company. The reduced visibility stems f…
OpenAI has begun rolling out GPT-6 Astra, its latest ChatGPT model, to organizations in its Daybreak program, with paid subscribers on Plus, Pro, Business, and Enterprise plans to gain access in the coming days. OpenAI c…
Containers limit blast radius but do not prevent autonomous coding agents from exfiltrating secrets, force-pushing to protected branches, or executing irreversible commands, according to a technical analysis of gating ag…