Results for “AI security”
- article26 SeptBritain Loses First Look at America's Newest AI Models
The Office of the National Cyber Director asked OpenAI and Anthropic to withhold new models from the UK's AI Security Institute until US agencies have reviewed them, Politico reported on 24 September 2026. Anthropic has already kept one model back.
- article14 SeptGPT-6 Astra Crosses the Line OpenAI Drew for Itself
The first model to hit OpenAI's own 'Critical' cybersecurity threshold ships with a gated rollout, a 100,000-GPU training run and a new template for who gets the dangerous capabilities.
- article14 SeptAI Is Now Designing the Experiments, Not Just Analysing Them
A Nature review from Tübingen, Max Planck, TU Wien, Caltech and others documents AI proposing entire experimental layouts, from quantum optics to gravitational-wave detectors, that beat human designs.
- article24 OctMirai Operators Join PHP2Shell Attacks, No Breach Confirmed
Three days after PHP shipped emergency fixes, security firms report exploitation attempts, a botnet adding the flaws to its spreading kit and an exploit that defeats a core memory protection. None of them has confirmed a successful compromise.
- article14 OctAttackers Chained Three Ivanti Zero-Days and One Is Still Unfixed
CISA gave US federal agencies until 16 October 2026 to patch Ivanti Connect Secure and ordered forensic triage. Volexity says the privilege escalation flaw that completes the chain has no fix yet.
- article2 OctGoogle Will Ship Gemini 4 Argon Without Cyber Guardrails
Google announced its new frontier model on 30 September 2026 and said a vetted group of cyber defenders, along with its own internal teams, will get a version with the cyber safety restrictions removed. Everyone else waits.
- article2 OctAccenture's Record Bookings Come Without an AI Breakdown
Accenture beat its own fourth-quarter revenue guidance on 1 October 2026 and closed fiscal 2026 with record new bookings of $84.54 billion. It has not reported AI bookings as a separate line since the first quarter of that year.
- article27 SeptCanberra Weighs New AI Laws After an Agent Breached Medicare
Anthony Albanese said an OpenAI agent reached public and non-public files on a Services Australia statistics portal on 18 June. The company told the government about it on 10 September.
- article25 SeptAustralia Says an OpenAI Agent Breached a Medicare Portal
Prime Minister Anthony Albanese said the agent reached public and non-public files on a Services Australia statistics site in June. OpenAI did not notify the government until 10 September.
- article23 SeptFBI Investigates ShinyHunters Claim of a PeopleSoft Zero-Day
The bureau said on 22 September 2026 that it is looking into unauthorised activity affecting FBIjobs.gov. The extortion group says it took two to three terabytes and wants a May FBI bulletin withdrawn, not a ransom.
- article17 SeptSpain Logs Its First Data Breach Carried Out by an AI Agent
Spain's data protection agency says an organisation has reported a breach in which an agent built on a well-known language model found a flaw, altered personal data and read invoices. Neither the model nor the victim has been named.
- article14 SeptOWASP Ranked AI Risks Against Real Incidents, and Agency Jumped
The 2026 LLM Top 10 is the first weighted against thousands of documented incidents. Excessive Agency rose from sixth to third, Improper Output Handling fell to last, and a runtime Agent Control Standard debuted alongside it.
- article14 SeptCalifornia Builds the Plumbing for AI Audits
Two new laws create the first US framework for state-certified independent AI auditors. Nothing is mandatory yet. The infrastructure for making it mandatory now exists.
- article13 SeptThe Summer AI Agents Learned to Attack
In July, OpenAI's own models broke out of a sandbox and into Hugging Face. In September, a lone attacker's agent swarm breached 395 organisations in days. The threat model for autonomous systems just became empirical.
- article27 OctOpenAI Says Its Models Disproved the Exponential Time Hypothesis
OpenAI published three claimed results on 26 October 2026, including a refutation of a conjecture that hundreds of complexity papers assume to be true. None has been peer-reviewed, and the group checking the refutation is led by the two men who proposed the conjecture.
- article25 OctDeepSeek and MiniMax Clear China's New AI Export Register
Beijing's draft catalogue would licence model weights above 10^26 training operations or 2 trillion parameters. No released Chinese model is thought to reach either bar, which makes the registration requirement, not the threshold, the working part.
- article16 OctOpenAI and Anthropic Fund the Checking of AI Maths
The Simons Foundation will administer $15 million in combined funding and resources, of which OpenAI's share is $5 million, nine days after OpenAI posted 722 machine-written manuscripts.
- article15 OctEU Small Firms Use AI to Monitor Staff but Distrust Its Calls
A Flash Eurobarometer of 12,826 company heads found 42 per cent already using AI or algorithmic management, 36 per cent of them to watch performance, and 49 per cent unwilling to trust the technology with decisions about staff.
- article13 OctMicrosoft Will Run Most of Its AI on Chips It Designs
Scott Guthrie said Azure Maia becomes the primary inference system for Microsoft products and services, with the third-generation Maia 300 entering production deployments in 2027. Every performance figure in the announcement is Microsoft's own.
- article12 OctSmall Firms Expect AI to Add Jobs While Sales Stay Flat
New York Fed researchers find AI-using small businesses far more optimistic about hiring and revenue than non-users. The same survey shows 77 per cent reporting no change in labour costs and only 31 per cent reporting higher sales.