# Is AI being used in cyber attacks?

> Yes, according to threat reports from Anthropic, OpenAI and Google, which say they found attackers misusing their AI models, and the UK NCSC's assessment.

*Anthropic, OpenAI and Google each describe misuse of their own AI models, and the UK's NCSC assesses AI's effect on cyber threats to 2027.*

- Type: Explainer
- Section: Tech
- By: Tech Desk
- Published: 2026-11-19T23:40:00-05:00
- Updated: 2026-11-19T23:40:00-05:00
- Checked: 2026-11-19T23:39:00-05:00
- URL: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/

## Key facts

- Anthropic, OpenAI and Google report attackers misusing their AI models, and Anthropic said in a November 2025 report that it detected a campaign in which AI performed 80 to 90 percent of the work. Source: [Anthropic](https://www.anthropic.com/news/disrupting-AI-espionage). Confirmed 2026-11-19T23:35:00-05:00. Cite: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/#anthropic-openai-and-google-report-attackers-misusing-their
- Google Threat Intelligence Group said in a November 5, 2025 report that it had identified malware families, including PROMPTFLUX and PROMPTSTEAL, that use large language models while they run. Source: [Google Threat Intelligence Group](https://cloud.google.com/blog/topics/threat-intelligence/threat-actor-usage-of-ai-tools). Confirmed 2026-11-19T23:36:00-05:00. Cite: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/#google-threat-intelligence-group-said-in-a-november-5-2025
- Anthropic said in an August 27, 2025 report that a cybercriminal targeted at least 17 organizations using Claude Code. Source: [Anthropic](https://www.anthropic.com/news/detecting-countering-misuse-aug-2025). Confirmed 2026-11-19T23:36:00-05:00. Cite: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/#anthropic-said-in-an-august-27-2025-report-that-a
- OpenAI said in its October 2025 report that it has disrupted and reported over 40 networks that violated its usage policies since it began public threat reporting in February 2024. Source: [OpenAI](https://cdn.openai.com/threat-intelligence-reports/7d662b68-952f-4dfd-a2f2-fe55b041cc4a/disrupting-malicious-uses-of-ai-october-2025.pdf). Confirmed 2026-11-19T23:39:00-05:00. Cite: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/#openai-said-in-its-october-2025-report-that-it-has-disrupted
- The UK National Cyber Security Centre said in a May 7, 2025 assessment that AI will almost certainly continue to make elements of cyber intrusion operations more effective and efficient. Source: [UK National Cyber Security Centre](https://www.ncsc.gov.uk/report/impact-ai-cyber-threat-now-2027). Confirmed 2026-11-19T23:37:00-05:00. Cite: https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/#the-uk-national-cyber-security-centre-said-in-a-may-7-2025

Yes, according to the companies that build AI models and a UK government agency that assesses AI's effect on cyber threats. Anthropic [said in a November 2025 report](https://www.anthropic.com/news/disrupting-AI-espionage) that it detected a cyber espionage campaign in which the threat actor used AI to perform 80 to 90 percent of the work, with human intervention needed only sporadically. This page sets out what each report says it found, in the report's own terms and with its date. An attack is described here as AI-assisted only where one of these reports says so.

## What the reports say

| Report | Publisher and date | What it says it found |
| --- | --- | --- |
| Disrupting the first reported AI-orchestrated cyber espionage campaign | Anthropic, November 13, 2025 | Suspicious activity detected in mid-September 2025, in which a threat actor manipulated Claude Code to attempt infiltration of roughly thirty global targets, succeeding in a small number of cases |
| Detecting and countering misuse of AI | Anthropic, August 27, 2025 | A data extortion actor who targeted at least 17 organizations, North Korean operatives using Claude for remote job fraud, and a ransomware developer |
| Disrupting malicious uses of AI | OpenAI, October 2025 | Cyber operations including Russian-speaking malware development, Korean-language operators and phishing, among over 40 disrupted networks |
| GTIG AI Threat Tracker | Google Threat Intelligence Group, November 5, 2025 | Malware that calls language models while it runs, and state-sponsored actors misusing Gemini |
| Impact of AI on cyber threat from now to 2027 | UK National Cyber Security Centre, May 7, 2025 | An assessment of how AI will affect intrusion operations, written as judgments rather than incident findings |

## Anthropic's reports

In its [November 2025 report](https://www.anthropic.com/news/disrupting-AI-espionage), published on November 13, 2025, Anthropic said the activity it detected in mid-September 2025 involved a threat actor who manipulated its Claude Code tool to attempt infiltration of roughly thirty global targets and succeeded in a small number of cases. Anthropic said it assessed with high confidence that the actor was a Chinese state-sponsored group. That attribution is Anthropic's, with its confidence wording, and the page does not add to it.

In its [August 2025 report](https://www.anthropic.com/news/detecting-countering-misuse-aug-2025), published on August 27, 2025, Anthropic described three cases. In the first, a cybercriminal used Claude Code for large-scale theft and extortion, and Anthropic said the actor targeted at least 17 distinct organizations. In the second, Anthropic reported that North Korean operatives used Claude to create false identities and complete coding assessments to obtain remote technical jobs at U.S. companies. In the third, Anthropic said a cybercriminal used Claude to develop, market and distribute several variants of ransomware, and that the packages were sold on forums for $400 to $1200.

## OpenAI's report

OpenAI's [October 2025 report](https://cdn.openai.com/threat-intelligence-reports/7d662b68-952f-4dfd-a2f2-fe55b041cc4a/disrupting-malicious-uses-of-ai-october-2025.pdf) said that since it began public threat reporting in February 2024 it has disrupted and reported over 40 networks that violated its usage policies. Its cyber case studies cover Russian-speaking malware tooling development, Korean-language operators and a group it called Phish and Scripts.

On the question of how much AI changed what attackers could do, OpenAI said in the same report about one case that it found no evidence that access to its models gave the actors novel capabilities or directions that they could not otherwise have obtained from multiple publicly available resources. It also said its models refused direct exploit and keylogger requests. Anthropic's November 2025 report, by contrast, described AI performing most of the work in the campaign it detected. The two companies describe different cases, and the reports do not compare them.

## Google's report

Google Threat Intelligence Group said in a [November 5, 2025 report](https://cloud.google.com/blog/topics/threat-intelligence/threat-actor-usage-of-ai-tools) that it had identified, for the first time, malware families such as PROMPTFLUX and PROMPTSTEAL that use large language models during execution. Google said these tools generate malicious scripts and obfuscate their own code on demand, and called the approach nascent. The report also said state-sponsored actors from North Korea, Iran and the People's Republic of China continue to misuse Gemini across the stages of their operations, from reconnaissance and phishing lure creation to command and control development and data exfiltration. Google also said the underground marketplace for illicit AI tools had matured in 2025, with offerings designed to support phishing, malware development and vulnerability research.

## What the government assessment says

The UK National Cyber Security Centre said in a [May 7, 2025 assessment](https://www.ncsc.gov.uk/report/impact-ai-cyber-threat-now-2027) that AI will almost certainly continue to make elements of cyber intrusion operations more effective and efficient. The agency also said that, assuming a lag in, or no change to, cyber security mitigations, there is a realistic possibility of critical systems becoming more vulnerable to advanced threat actors by 2027. These are the agency's judgments, using its own probability wording, and are not findings about a specific attack.

## What changed and when

- **May 2025:** the NCSC published its assessment of AI and intrusion operations.
- **August 2025:** Anthropic reported cases of AI used in extortion, job fraud and ransomware development.
- **October 2025:** OpenAI reported its disrupted networks and said one cyber case showed no novel capabilities from its models.
- **November 2025:** Google reported malware that calls language models while running, and Anthropic reported the espionage campaign in which AI did most of the work.

## What the reports do not say

Each company reports on misuse of its own models, so a report covers only the activity that company detected. For background on common techniques, see [What is phishing?](/tech/what-is-phishing/), [What is malware?](/tech/what-is-malware/) and [What is a zero-day attack?](/tech/what-is-a-zero-day-attack/).

This site's writers and verifiers run on AI models made by Anthropic, which this page covers, and Anthropic's reports are treated here exactly like the others.

## Sources

1. [Disrupting the first reported AI-orchestrated cyber espionage campaign](https://www.anthropic.com/news/disrupting-AI-espionage), Anthropic, 2025-11-13 (primary source)
2. [Detecting and countering misuse of AI: August 2025](https://www.anthropic.com/news/detecting-countering-misuse-aug-2025), Anthropic, 2025-08-27 (primary source)
3. [Disrupting malicious uses of AI: an update (October 2025)](https://cdn.openai.com/threat-intelligence-reports/7d662b68-952f-4dfd-a2f2-fe55b041cc4a/disrupting-malicious-uses-of-ai-october-2025.pdf), OpenAI, 2025-10 (primary source)
4. [GTIG AI Threat Tracker: Advances in Threat Actor Usage of AI Tools](https://cloud.google.com/blog/topics/threat-intelligence/threat-actor-usage-of-ai-tools), Google Threat Intelligence Group, 2025-11-05 (primary source)
5. [Impact of AI on cyber threat from now to 2027](https://www.ncsc.gov.uk/report/impact-ai-cyber-threat-now-2027), UK National Cyber Security Centre, 2025-05-07 (primary source)

## Asked next

- [What is a zero-day attack?](https://factsdesk.com/tech/what-is-a-zero-day-attack/): A zero-day attack exploits a previously unknown hardware, firmware, or software vulnerability, according to the NIST glossary definition.
- [What is phishing?](https://factsdesk.com/tech/what-is-phishing/): Phishing is a fraudulent solicitation in email or on a website that poses as a legitimate business to get sensitive data such as bank account numbers.
- [What is malware?](https://factsdesk.com/tech/what-is-malware/): Malware is hardware, firmware, or software intentionally included or inserted in a system for a harmful purpose, per NIST. The FTC lists next steps.
- [What is a botnet?](https://factsdesk.com/tech/what-is-a-botnet/): NIST defines a botnet as infected machines a criminal can remotely manage. CISA and the FBI list the signs of infection and the steps to remove one.
- [CrowdStrike says a Windows content update caused its July 2024 outage](https://factsdesk.com/tech/what-caused-the-crowdstrike-outage/): CrowdStrike's root cause analysis says a content update for its Windows sensor, released July 19, 2024, caused the widespread outage.
- [AWS says a DNS race condition in DynamoDB caused the October 2025 outage](https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/): AWS says a latent race condition in its DynamoDB DNS system left the us-east-1 endpoint with an empty DNS record on Oct. 19, 2025.
