No one can promise what 2027 holds, but the best current review says AI helps attackers at several steps while fully autonomous, end-to-end attacks have not been reported. The International AI Safety Report 2026 describes growing capability and real misuse, together with clear limits. Plan for faster attackers, not for a robot takeover.
The source
The International AI Safety Report is a review of the science on general-purpose AI capabilities and risks, led by Yoshua Bengio. The second edition, the 2026 report, came out on 3 February 2026. According to its arXiv abstract, over 100 experts contributed, and 29 nations, the UN, the OECD and the EU each nominated a representative to its advisory panel.
What it says on cyber
As summarized by a trade press report on the document, AI is taking over more pieces of attack workflows, such as spotting targets and weaknesses and producing malicious code. But it says AI systems cannot reliably execute long, multi-stage attack sequences without humans. Observed problems include agents issuing unrelated commands, losing track of where they are in an operation, and failing to recover from basic errors without human help. The report states that fully autonomous, end-to-end attacks have not been reported.
The report's own site adds a caution about defences: a November 2025 update says sophisticated attackers can often bypass current safeguards and that the real-world effectiveness of many safeguards is uncertain.
What to take from it
- Speed, not magic. Expect faster reconnaissance, more convincing phishing text and quicker exploit development.
- Humans stay in the loop. The attacks still need operators, which means normal detection and response still apply.
- Measurement is hard. Benchmarks can mislead, so read the full report for specifics before quoting numbers.
What to do about it
- Shorten patch time for internet-facing systems.
- Use phishing-resistant authentication.
- Reduce the time between a new advisory and your fix.
- Review how your own AI agents are permissioned, since they are also a target.
See Cyber Pearl Harbor: how to read scary headlines for the framing.
What would change the picture
Watch for evidence rather than headlines. The report's own caution is that attackers' use of AI is hard to attribute and that benchmarks can mislead. Reliable signs of a step change would be documented, fully autonomous attack chains in incident reports, not demonstrations in a lab. Until a reputable source reports such cases, treat claims of imminent autonomous mass attacks as unproven.
Meanwhile the defender's side improves too. The same tools help with code review, triage and detection. Teams that adopt them with care and keep humans accountable for decisions will be better placed than teams that ignore them.
Quick check
Have fully autonomous AI cyberattacks been reported?
The 2026 International AI Safety Report, as summarized by trade press, says fully autonomous, end-to-end attacks have not been reported.
Sources
- International AI Safety Report
- arXiv: International AI Safety Report 2026
- TechInformed: coverage of the 2026 report on AI and end-to-end cyberattacks
Last reviewed: 19 October 2026.
