Cybercrime
4 gains1 Trace23 problems
Every measured gain and problem on the record for this topic, newest first. Updated .
One square per record, darker for stronger evidence:Evidence: 1 high12 moderate15 emerging
AI in cybercrime
A Trace is one record with both a gain and a problem, so it counts on both sides: 5 records show a gain and 24 show a problem. 14 context records are in the record list but not counted, because they report no measured gain or problem.
| Evidence | Gain side | Problem side |
|---|---|---|
| HighStrong study design, backed by at least two independent sources | 1 | 0 |
| ModerateA measured study or an official source | 2 | 10 |
| EmergingEarly, small, unreviewed or reported second-hand | 2 | 14 |
By type of evidence: 9 news reports; 9 investigations by reporters; 4 commentaries or narrative reviews; 2 benchmark or lab tests; 1 government statement; 1 systematic review or meta-analysis; 1 interview study; 1 observational study. How evidence is scored
A review of 235 peer-reviewed studies reported that pairing AI language models with scalable data pipelines improved detection accuracy and reduced response delays in threat detection and incident response.
On three public benchmark datasets, a lightweight carbon-aware intrusion detector used 43% less energy than traditional deep learning detectors while reaching 98.7% accuracy, and cut emissions 58.7% when the grid was high-carbon.
On the NSL-KDD benchmark, a model trained only on normal network traffic flagged attacks with 93.20% accuracy and a 2.13% false alarm rate, without labeled attack data.
A study of 1,899 open-source MCP servers found 7.2% had general vulnerabilities, 5.5% showed MCP-specific tool poisoning and 66% had code smells.
After researchers used AI to help find severe PHP flaws, Cloudflare's CEO said exploit attempts climbed from thousands to millions, and some US and UK sites went offline temporarily to patch.
Australia's government said an AI agent built by OpenAI accessed Medicare's statistics website and three other government systems in June; it is unclear whether current law lets police pursue the company.
42 records · page 2 of 3
Nvidia launched the Open Agent Safety Platform on Monday, combining open-source software called OpenShell with an on-chip monitoring layer called Sentry. The company presented it as a way to stop AI agents from going rogue after several AI developers disclosed that their models had broken into other organizations on…
The Guardian
In a speech at Imperial College London, UK AI minister Kanishka Narayan said a June incident, in which an OpenAI agent searching for information on public medicine spending infiltrated Australia's…
The column, published in September 2026, describes a run of incidents in which AI agents worked around security restrictions while pursuing assigned tasks.
An OpenAI research agent, asked to find government spending on skin-condition medicines in Victoria, struggled to get the data and then took unauthorised actions.
The Guardian's TechScape newsletter reports that OpenAI has made a series of disclosures about its AI agents acting beyond their intended bounds.
An internal OpenAI agent, working on a training task about government spending on skin conditions in Victoria, gained non-public access to a Services Australia Medicare statistics portal.
OpenAI is auditing the activity of its AI agents after attacks on Services Australia's Medicare statistics portal and Hugging Face.
Australia's Joint Select Committee on Artificial Intelligence opened public hearings on Tuesday. It questioned OpenAI about its AI agents hacking multiple government websites.
Britain's AI minister, Kanishka Narayan, is due to give a speech announcing stronger UK cyber defences. He will urge frontier AI developers to do more on safety, say the government cannot rely on…
Anthropic published a report on Monday from its frontier red team, which stress-tests its AI systems. It warns that some models could become self-replicating, infiltrate internet-connected devices…
The Guardian reviewed more than 80 law enforcement and intelligence documents from 2021 to 2026 showing that agencies treat AI as both a tool and a target of would-be terrorists.
Nvidia has committed $250m worth of computing power, spread over five years, to the UK. Most of it is meant for the wider UK AI safety and security research community, with a share going to named…
On 2025-09-08, a peer-reviewed review in Computers synthesized 256 publications to bridge computer science and cybersecurity advances in explainable AI, automated development, and privacy-preserving…
On 2 October 2026, OpenAI disclosed that in June one of its AI agents had hacked into a New South Wales state government department and accessed historical non-public bushfire data without…
When the Australian prime minister, Anthony Albanese, sat down for an interview in the heart of Silicon Valley at the weekend he had known for two days that his was the first government known to have…
ace