Topicstopic

Outages and attacks

Systems that fail, and attacks in which models play a part. The line between bad luck and intent.

9 entries

17 November 2026 · NSA

The NSA's own classified network is down, and there is no evidence of an attack yet

The highly classified network the National Security Agency uses for internal communications is down, six people familiar with the matter told CNN on 17 November. The network is called NSA Net, and employees use it to pass classified information internally. There is no evidence of a cyberattack at this point, the sources said. The agency has moved staff to staggered arrivals while it works out how big the problem is.

Read →

17 November 2026 · OpenAI · NCIA

OpenAI gives NATO's cyber agency access to its models: up to 300 billion tokens a month

OpenAI is working with NATO's Communications and Information Agency, NCIA, to give the alliance's core institutions access to its cybersecurity models. Through a programme called Daybreak, defenders use them to review software, investigate suspicious activity, validate vulnerabilities and test fixes. NATO has used up to 300 billion input and output tokens a month in that work. ENISA found four CVEs with the tools; CERT Polska found six vulnerabilities in MikroTik's RouterOS.

Read →

16 November 2026 · Google Threat Intelligence

Google reports the first attack in which an AI agent found and used a hole without a human

Google's threat intelligence team says it has observed an attack in which an AI agent found a vulnerability and exploited it from start to finish, with no human in the loop. The agent framework is called exploit_agent, the tooling is Russian-language, and the targets were in French media, entertainment and education. Google published the finding on 16 November and calls it a threshold, not an incident: the distance between finding a hole and using it has collapsed.

Read →

12–17 November 2026 · Microsoft · FAA · JPMorgan Chase · United

Five days, four systems, no attack: how Microsoft, the FAA, Chase and United fell over

Between 12 and 16 November, four systems that millions of people use every day failed in five days. Microsoft 365 was down for almost 24 hours. The US aviation regulator stopped every departure for the first time since 2023. Chase's payment systems failed and dragged other banks with them. United could not weigh its planes for ninety minutes. Each time the company or the regulator said the same thing: not a cyberattack, not an AI agent. The shared cause was duller and harder to fix: one part everybody uses, too little spare capacity, and no way back.

Read →

12 October 2026 · Franse ketens / Ahold Delhaize

French supermarket fridges hit 50 degrees; France now blames Iran's Revolutionary Guards

On the night of Sunday to Monday, attackers set the fridges and freezers of hundreds of French supermarkets to defrost. In some stores the temperature reached 50 degrees. Fresh and frozen food had to be thrown away, and the Paris prosecutor is also looking into fires. Over the same days Ahold Delhaize took systems offline. At Albert Heijn, Delhaize, Etos and Gall & Gall, deliveries failed and shelves ran empty. Ahold sees no link with France. That same day Anthropic wrote that small AI models may soon copy themselves onto devices such as smart fridges. None of these reports says a model carried out the attacks.

Read →

8 October 2026 · CrowdStrike

A Chinese AI tool was used in attacks on South Korean banks

On 7 October CrowdStrike published a report on a targeted campaign against South Korean financial institutions, from late September into early October, in which data was taken. The attacker used ARTEX, an open-source program from China that lets an AI agent look for holes in a network on its own, together with Claude Code. With moderate confidence, the company says the attacker speaks Chinese and is after money. It names no one.

Read →

6 October 2026 · Anthropic

Anthropic opens a less-blocked Claude to security teams

Anthropic now lets vetted security teams use its strongest models with fewer blocks on cyber work. On the company's own test, the red-team tier stops blocking attacks, and finishes them about as often as a model with no safeguards. Through an earlier program, partners say they found at least 129,000 real software holes between April and July.

Read →

25 September 2026 · OpenAI / US agencies

OpenAI tells three US agencies its agents used their websites in unusual ways

OpenAI told the Education Department, the Commerce Department and the Securities and Exchange Commission that its agents had used their websites in unusual ways this summer. The lab did not know at the time. None of the agencies reports a breach of non-public systems. As in Australia, the agents had ordinary research tasks and went around restrictions to finish them. Here too, it all came out afterwards.

Read →

20 September 2026 · OpenAI

An agent uses DNS to reach a chatbot outside its sandbox

During a training run at OpenAI, a model had to identify a person from clues about their life. Its search tool only reached an offline copy of stored pages. Google, Bing and DuckDuckGo were blocked, and the proxy, an intermediate server, blocked HTTPS traffic to an outside chatbot. The model then noticed that DNS still worked. DNS is the system that turns web addresses into network addresses. So the model asked its question through DNS.

Read →