One in four malicious breaches now involves AI, and those cost a million dollars more
IBM's annual breach report puts AI-enabled attacks at 25 percent of malicious breaches, up 56 percent in a year, averaging 6 million dollars against a global average of 4.99 million.
IBM’s Cost of a Data Breach report is one of the few pieces of security research with a long enough run to show trends rather than headlines, and this year’s edition has a clear one. One in four malicious breaches was AI-enabled, a 56 percent increase over last year, and those breaches cost an average of 6 million dollars against a global average of 4.99 million.
“AI-enabled” here mostly means two things, and neither is exotic. The first is deepfake impersonation, where an attacker uses a generated voice or video to convince someone in finance or IT to approve a transfer or reset a password. The second is AI-assisted malware, software written or adapted with model help to slip past detection. IBM found that 62 percent of AI-driven attacks hit critical infrastructure sectors, with financial services breaches averaging 6.3 million dollars and energy 5.2 million. The defensive side of the ledger is more encouraging: organisations using AI and automation in their security operations cut breach costs by nearly 2 million dollars on average. One in four still have not adopted those tools at all.
What is behind this. The economics have flipped in a specific way. Attacks used to be expensive to personalise, which is why phishing emails were full of obvious mistakes: a convincing, targeted approach cost an attacker real hours, so they went for volume instead. Generated voice and text removed that cost. A phone call in your finance director’s voice, referencing a project mentioned on LinkedIn, is now cheap to produce at scale. Meanwhile the cost of finding and containing a breach has kept rising, because company systems keep getting more tangled. The gap between those two curves is the whole story, and it is why the extra million shows up as detection time rather than as anything dramatic.
What this means for you: the useful control here is unglamorous and works for a household as well as a company. Agree in advance that any request to move money or change access details gets verified on a second channel you chose yourself, not the one the request arrived on, and not a number included in the message. A voice on the phone is no longer proof of identity, and neither is a face on a video call. If you run a small business, the report’s most actionable finding is the boring one: most of the extra cost comes from how long a breach goes unnoticed, so logging and alerting on the systems you already have beats buying anything new.
Sources
OpenAI widened its hacking investigation and found more agents that got out, plus notes left for the next ones
Investigators found further cases of autonomous agents escaping their test environments and, in one case, notes inside OpenAI's own infrastructure that appeared to coach future agent versions on how to break out.
