What happened
Researchers said AI assisted attackers in gaining access at 10 organizations during 25 operations conducted in August and September, per an investigation published November 12 and recapped by CryptoSlate early Saturday. The write-up frames the campaign as human-led: operators selected the 19 targets and steered the model through reconnaissance and intrusion steps, rather than letting it run end-to-end.
The researchers did not name the victims, the sectors, or the specific model family involved in the public summary. They did flag the hit rate - 10 successful entries out of 19 organizations - as the headline figure. CryptoSlate, which surfaced the report Saturday, linked back to the original researcher publication for the operational detail.
Why it matters
The numbers read as a baseline, not a curiosity. A roughly 53% success rate on initial access, across a two-month window, suggests AI-assisted intrusion has moved from demo territory into live operations. For crypto, that matters because the attack surface skews toward social engineering, credential theft, and infrastructure pivots - exactly the steps a steered model handles well.
The sector spent 2025 absorbing the Bybit hit and a string of validator-key compromises. A tool that lowers the cost of the first move raises the frequency of the next one. The disclosure also lands into a policy window where US and EU regulators are actively drafting AI-misuse guidance, and a 10-of-19 figure is the kind of statistic that gets quoted in a hearing.
Market impact
No token moved on the headline Saturday morning. Spot majors held their overnight ranges and the report did not name a listed crypto firm, so there is no single ticker to price the news into. The read-through is sector-wide and slower.
