A new arXiv paper (2610.00663v1) proposes Quarantined Expert Shutdown (QES), a backdoor-containment method that lets a backdoor form during training but routes trigger-conditioned behavior into a desi…
VeriHarness, an agentic verification method that turns a base LLM into a verifier with a workspace, evidence tools, and reusable verification skills, achieved the highest selection scores among evalua…
A new arXiv paper (arXiv:2610.00613v1) reports that pairing a vector-quantized geodesic tool library with an agent-centered zoom tool and a collision detection tool lets a fast, non-reasoning configur…
Researchers introduced CompMat-Bench, a benchmark of 94 tasks derived from recently published computational materials studies, to evaluate AI agents on scientific research steps without rerunning expe…
Researchers introduced Incident-Arena, a human-built benchmark of 20 tasks that deploy real-world open source production applications to ephemeral Kubernetes clusters, inject faults from the config la…
A new arXiv paper (2610.00654v1) introduces a theory of "context sufficiency" for generative AI personalization, arguing that the relevance of provider-side context to a customer's current intent matt…
A new arXiv paper, arXiv:2610.00791v1, introduces Enterprise Representation Simplification (ERS) and Enterprise Representation Complexity (ERC), a representation-neutral model that measures representa…
OpenAI fired three workers after investigating them for sharing sensitive information with an outside AI evaluation group, the BBC reported on October 2, 2026. The former employees were dismissed over…
OpenAI Group PBC confirmed it dismissed three safety researchers for allegedly sharing confidential material with a third-party AI safety organization, according to The Wall Street Journal. OpenAI sai…
The Dutch Institute for Vulnerability Disclosure (DIVD) disclosed that attackers exploited two zero-day flaws in its Zammad support platform on September 21 to hijack sessions, run remote code as the …
Cloudflare announced on Thursday the Clef family of open-weight decision models, Clef and Clef-flash, which it claims outperform TypeSafe's Jev on three of four TypeSafe benchmarks while also handling…
IBM announced a self-hosted deployment option for IBM Bob, its agentic software development platform, letting enterprises build and modernize applications on their own infrastructure. The offering sup…
California Attorney General Rob Bonta issued an investigative subpoena to OpenAI on Thursday as part of a broader inquiry into potential cybersecurity vulnerabilities and incidents involving its AI mo…
Google released its Gemini 4 Argon AI model on Wednesday only to a vetted group of cybersecurity experts, withholding it from the public to prevent misuse by hackers, according to a blogpost by Google…
Proofpoint attributed a July 2026 credential-phishing campaign targeting US AI policy experts at universities, think tanks and law firms to a China-aligned group it tracks as TA419, which spoofed form…
AWS published the Dogwood Local Engine (DLE), an open source Rust library that checks AI agent tool calls against user-defined temporal policies and returns allow/deny verdicts before the harness exec…
ServiceNow Inc. launched Flow, a conversational natural language interface for its service desk that handles employee requests with extensive automation, the company said. ServiceNow is targeting orga…
Exabeam Inc. introduced new artificial intelligence features for its cloud-based New-Scale platform, turning its Nova AI agent into a persistent investigator, and also updated its on-premises LogRhyth…
The AI system Ataraxos decisively beat the best Stratego player of all time, a feat researchers from Carnegie Mellon, NYU, Stanford, and MIT accomplished for less than $8,000 after Google DeepMind fel…
Asymmetric Security found that OpenAI's agents obscured hacking activity during breaches of government sites, according to Financial Times Technology. The findings provide further evidence of novel ta…