AI & Tech
Google says attackers used AI alone to find and exploit software flaws
Google's threat intelligence unit says a tool found on an exposed server ran the full cycle of finding and exploiting flaws with no human involved. Victim data points to French media, entertainment and education websites.
AI text based on reports by The Next Web, ZDNet, adevarul.ro and other publishers. Published . How it is made
For membersThe first 30 seconds for everyone. Synthetic voice (AI).
In brief
Google says a tool called exploit_agent, found on an exposed staging server, compares patches, finds flaws and writes and runs exploits without a human, a first for the group.
Huntress built working PaperCut exploits within a day of disclosure, and GreyNoise linked an AI-driven campaign against hundreds of organisations to a date four days after disclosure.
The FBI says Artex-linked attacks can reach active compromise in under three hours, and at least 68,000 people had personal data stolen from South Korean financial institutions.
Microsoft, which says the hackers are likely Chinese, reported access to the Microsoft 365 cloud environments of at least 2,250 ConnectWise customers, a finding the FBI confirmed.
Google's Q4 AI Threat Tracker says AI token batches sell for as little as $0.10 through LLM Mixers, while victims face usage bills over $100,000.
Xint.io found GPT 6 Sol located a V8 flaw in 13 minutes for $4.24 in September, down from $22 and 44 minutes for GPT 5.6 Sol in July.
- February 2026Best available models fail to find the bug after 25 hours and $350 of scanning
- July 2026GPT 5.6 Sol finds the flaw in 44 minutes for $22
- September 2026GPT 6 Sol finds the flaw in 13 minutes for $4.24 on its first guided prompt
An autonomous tool and its targets
Google's Threat Intelligence Group and Mandiant said on Monday, 16 November, that attackers had used AI to find software flaws and exploit them with no human involved, something the group said it had not seen before. The finding appears in the group's latest AI threat report. Investigators located the tool, which they call exploit_agent, on an exposed staging server. It works with several frontier and locally run AI models, compares popular web software before and after each patch, flags likely flaws, and then writes and runs exploits on its own. The previous quarterly edition, issued two months earlier, had said the group had not yet seen fully autonomous attack pipelines used against targets.
The same server held data stolen from victims, along with scripts for brute-forcing passwords, installing webshells and removing webshells planted by rival attackers. A configuration file carried comments in Russian. Based on victim data, the operator appears to have targeted French media, entertainment and education websites. The report does not establish that the operators acted on the orders of Russian authorities. Google described the shift in direct terms.
“fully automate vulnerability discovery and exploitation without a human in the loop for the first time”
From patch to exploit in days
Google argues that patched flaws are easier for AI to attack than unknown ones, because attackers have a limited body of material to study: code before and after changes, release notes and published research. The report points to two PaperCut print server flaws that attackers chained together in late August. Researchers and the security firm Huntress used AI to build working exploits within a day of disclosure. GreyNoise, as cited by Google, said an attacker likely used AI four days after disclosure to run a campaign against hundreds of organisations. After PHP flaws were patched in October, hundreds of proof-of-concept exploits appeared that did not work, and the report says AI likely wrote many of them.
Open-source frameworks and the FBI's Artex warning
Google said several groups now use open-source AI hacking frameworks in live attacks, including financially motivated gangs and suspected Chinese spy groups. It saw the CAIRN framework used against targets including the US aerospace sector, and it expects attackers to rely less on humans and become stealthier as the tools improve. In a flash cybersecurity advisory, the FBI warned that hackers were using an open-source tool called Artex against financial services, government, tech, defense, phone and internet companies, and law firms. The agency said Artex intrusions can be noisy, yet the speed they allow makes it far harder for human defenders to keep up. In some cases Artex-linked attacks go from targeting a system to active compromise 'in under three hours'. Artex has been used since its first public debut in July, including against South Korean financial institutions where at least 68,000 people had their personal data stolen. The FBI also confirmed earlier Microsoft findings that hackers used Artex against ConnectWise and reached the Microsoft 365 cloud environments of at least 2,250 of its customers, and Microsoft says the hackers are likely Chinese.
Cheaper access and cheaper zero-day discovery
Google's latest tracker found criminals reselling stolen enterprise AI access through 'LLM Mixers', underground storefronts where AI token batches sell for as little as $0.10 while the organisations whose accounts were hijacked face usage bills above $100,000. The report describes LLMjacking as the theft of AI credentials such as usernames, passwords and API keys. John Hultquist, chief analyst of the Google Threat Intelligence Group, said the group had tracked LLMjacking rising through 2026. Separately, Xint.io tested seven models launched since February 2026, two from OpenAI, three from Anthropic and two open-source models from DeepSeek and Z.ai, on a flaw in Google's V8 engine tracked as CVE-2026-102321, a type confusion bug that was absent from the models' training data. In February the best models then available failed after 25 hours and $350 of scanning. GPT 5.6 Sol found the flaw in July in 44 minutes for $22, and GPT 6 Sol found it in September in 13 minutes for $4.24 on its first guided prompt. Only three of the seven models found the bug: GPT 6 Sol, GPT 5.6 Sol and Claude Opus 5.
- GPT 5.6 Sol (July 2026)
- 22 USD
- GPT 6 Sol (September 2026)
- 4.24 USD
Data from the article text
Wider pressure on defenders
Governments and critical infrastructure operators are still dealing with serious flaws found in PHP, a language that underpins about 70% of the web. Authorities in Europe are also facing attacks on supermarket refrigeration systems, some of which have been partly blamed on Iran. The new development adds to concern about AI-driven attacks, a worry that has grown since Anthropic introduced its Mythos model in April. Google expects attackers to rely less on humans as the tools improve.
Ask about this text
Sources and context
- The Next WebGoogle says attackers used AI alone to turn patches into exploits
- ZDNet5 ways AI will change how you defend your business in 2027
- adevarul.roGoogle warns: Russian-speaking hackers launched the first fully automated cyberattack using artificial intelligenceGoogle avertizează: Hackeri vorbitori de rusă au lansat primul atac cibernetic complet automatizat cu ajutorul inteligenței artificiale · original in Romanian
- ΣΚΑΪGoogle: Russian-speaking hackers behind possibly the first fully automated cyberattack with AIGoogle: Ρωσόφωνοι χάκερ πίσω από πιθανώς την πρώτη πλήρως αυτοματοποιημένη κυβερνοεπίθεση με AI · original in Greek
Pollar is free and carries no ads in the text. Its members keep it going.
Support Pollar