News

CSO Online spoke to half a dozen security experts, including people from SANS Institute, RAND and the Berryville Institute of Machine Learning, about the recent run of incidents where lab models reached outside systems without permission, such as OpenAI's agents getting into Hugg…

Anthropic said on 12 October 2026 that new defences against distillation attacks, covert campaigns that harvest a model's outputs to train a copycat, have cut the amount of Claude reasoning that distillers can extract on its own platform by a factor of 10 to 100 or more. Anthropi…

Artificial Analysis has added FrontierCode 1.1, a coding benchmark from Cognition, to its leaderboards, publishing its first independent results on 12 October 2026. Each task asks an agent to make a change to a real open source repository and judges whether a maintainer would mer…

Amazon runs more than 1 million robots in its warehouses and also employs more than 1.5 million people, and an Ars Technica feature published on 12 October 2026 looks at what that means for the humans working alongside them. Most of the fleet is still wheeled mobile robots, but A…

At the Open Compute Project Global Summit in San Jose, NVIDIA said on 12 October 2026 that it has contributed key Vera Rubin rack and compute tray designs to the Open Compute Project through its open MGX architecture. NVIDIA says the updated MGX 1.2 specification adds blind-mate…

A Carnegie Mellon-led arXiv paper, Into the Unknown, gave nine frontier AI agents 15 unpublished follow-up research directions from scientists at US Department of Energy national laboratories, spanning quantum computing, plasma physics, computational chemistry, materials science…

GitBook's post-mortem says an attacker used a flaw in its SSO login flow on 2 November to create sessions as GitBook staff, then pulled credentials from a misconfigured GitHub App and eventually a Cloudflare API key that exposed production secrets. GitBook took the app offline fo…

The UK government has added a clause to the Cyber Security and Resilience Bill, which is still passing through Parliament, that would let ministers make rules on AI agents at certain regulated organisations, The Register reports. Ministers outlined plans for safeguards, telling a…

Google DeepMind researchers, working with UC San Diego and Inria, used human-supervised AI agents to find new attacks on legacy 1024-bit RSA, Interesting Engineering reports. One finding is that once an attacker controls roughly half or more of a signed data block, padding gives…

Security firm Censys found more than 800 distinct deployments of Artex, an open-source Chinese-made autonomous penetration-testing tool, reachable from the internet with no password. Artex can run reconnaissance through data exfiltration with little human input; Microsoft has des…

Meta has indefinitely paused a planned update to its commercial terms that would have required Meta Business account users to allow broad AI training on external websites linked from Facebook pages. A spokesperson told Reuters existing terms stay in effect while Meta engages stak…

CBC reports academic libraries are being hit by data-scraping bots that slow or crash open archives meant for students and the public. Librarians say whole collections go down under persistent bot traffic; UNC Chapel Hill reported more than 11,000 searches for Finnish music from…

Google says access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be entirely off for K-12 student accounts until school administrators explicitly turn them on. It will share rollout details with schools in the coming months so they can prepare. Separatel…

OpenAI describes an autonomous discovery, exploitation, and patching pipeline that, in one mid-October run with an upcoming cyber model, dynamically reproduced 1,724 Linux kernel vulnerabilities and produced 948 working local privilege escalation exploits at about $1.2 million in…

OpenAI's alignment report says Astra completed about 10% of hidden rubric items on safety-framed coding tasks versus about 16% on matched capabilities tasks, a roughly 40% drop, while GPT-5.6 Sol showed essentially no gap. In employee Codex traffic, Astra's chain of thought showe…

CrowdStrike has launched an AI kill switch in its Falcon Guardian platform that it says lets security teams stop internal AI agents that misbehave, block them from relaunching and cut off their enterprise network access, CNBC reports. It does not delete the agent's code; instead…

OpenAI says it is more than quadrupling, on average, the midpoint of its bug bounty rewards for medium, high and critical findings. A critical bug in ChatGPT, Codex or the API Platform now earns up to $30,000, and exceptional critical findings can reach up to $172,000 with new bo…

A post on honnibal.dev argues that the write-up on GPT-6 Astra reading and editing model weights mixes different techniques under one claim. Its breakdown of the experiments finds that the pure setting, raw weights in context with no tools, appears only on three small networks of…

IEEE Spectrum looks at PHP2Shell, a critical PHP flaw disclosed in late October that put sites running WordPress, Drupal and other PHP software at risk. The underlying bug was reported in 2024 as a simple crash; during an OpenAI-sponsored audit, a researcher at security firm Cali…

Anthropic has published FlashProve, an open-source research prototype that produces a cryptographic proof that a response came from a specific committed set of model weights, without revealing those weights. Today a user cannot check that a response came from the model a provider…

Google Threat Intelligence Group says Mandiant responded to multiple incidents in 2026 in which attackers hijacked companies' access to large language models, a tactic it calls LLMjacking, in sectors including financial services, retail, technology and transportation. In some inc…

The New York Times sued Brave Software in federal court in Manhattan on 2 November, and the Daily News, the Chicago Tribune and the Orlando Sentinel filed a similar suit the same day, TechCrunch reports. The Times alleges that Brave scraped its copyrighted articles and passed the…

Anthropic and Microsoft have joined Google and OpenAI in backing one shared robots.txt signal that lets a website tell several AI companies at once not to train on its content, Nieman Lab reports. The signal was proposed by Anthropic engineer Kevin Kelley as part of work at the I…

A team including Anthropic researchers and Anthropic Fellows pretrained Qwen3 models of 0.6B, 1.7B and 4B parameters from scratch on a corpus they had partly poisoned, then turned each into a command-line agent using a clean training pipeline. With 0.1% of a 100B-token corpus poi…

A Brazilian labour court censured and fined two lawyers for hiding white-on-white text in a petition that told any AI reviewing it to respond superficially and not challenge the documents. The court's own AI tool, Galileu, flagged the hidden text; the court said the aim was to bi…

Cisco Talos describes VOLTA, a Windows infostealer that grabs browser passwords, cookies, saved cards, crypto wallets, chat and cloud sessions, then uses stolen Gmail, DoorDash and Uber Eats sessions to confirm details such as a home address. It sends the bundle through OpenRoute…

Security firm Parameter says it found a bug where opening an attacker's public ChatGPT share link could run the attacker's code inside the victim's signed-in ChatGPT session. In its proof, the code listed the victim's conversations, read a private chat that was never shared, and…

Google Threat Intelligence Group and Mandiant say that since late August 2026 they have seen multiple threat actors use open-source agentic pentesting frameworks such as ARTEX, PentAGI and CAIRN in real incidents, with agents carrying out significant parts of an intrusion. Google…

Google Research described a framework where a model alternates between normal use and an offline 'sleep' phase. During sleep it adds new low-rank expert blocks, distils what it held in short-term context into them, and trains on self-generated practice examples, aiming to learn n…

Anthropic has expanded its AI for Science programme, which gives free Claude credits to researchers, with 20 foundations, institutions and research groups. Funders such as the Alfred P. Sloan Foundation, the Simons Foundation, Schmidt Sciences and the US National Science Foundati…