Interpol Says AI Isn't Reinventing Cyberattacks, Just Making Them Faster and Harder to Catch

Speaking on the sidelines of Tech Week Singapore, Interpol's global chief information security officer Bjorn R. Watne told CNBC that artificial intelligence is "an evolution and not a revolution" in cybercrime. His point was specific: AI isn't inventing new criminal methods, it's making old ones faster, more scalable, and harder to spot. Scammers can now hit multiple victims at once, lean on AI translation to fake fluency in any language, and use synthetic digital identities to blur the line between a real customer and a fraud.

Watne's advice to companies was blunt and practical. Identify your "crown jewels" first, he said, meaning the specific assets a business actually can't survive losing. Then figure out who wants them and why, and build defenses around that threat, not a generic checklist. "There is no need for everyone to try and protect against everything all at once when they're not being targeted by it," he said, according to CNBC.

His sharper warning was about agentic AI, systems that act on a user's behalf rather than just answering questions. Watne said the risk moves beyond bad information into bad actions, with physical consequences once AI runs things like vehicles. He also flagged a trust gap: people are more careful with a bank PIN than they are handing permissions to a new app, even though the app may now be capable of acting autonomously.

The hype problem

Two outlets covering the same Interpol remarks, TechBuzz.ai and Europe Says, ran near-identical write-ups built around warnings that should make a CISO's "blood run cold." Neither outlet quotes Watne saying anything that dramatic. Both cite claims about AI "swarms" and autonomous malware without naming a specific incident, contradicting the one clear thing Watne actually said on the record: this is an evolution, not a revolution.

Fox Business offered a more grounded version of the same debate. Yagub Rahimov, founder and CEO of Polygraf AI, told the outlet that AI is shifting cybercrime from lone hackers to coordinated "swarms" of autonomous agents that divide tasks and act as a group. "You cannot secure machine-speed activity with human-speed monitoring," Rahimov said. He also said the emergence of AI swarms doesn't necessarily mean a cyber doomsday is imminent, and pushed back on what he called alarmist coverage of recent AI incidents, saying "most of these news things are not as dangerous as what the news is trying to demonstrate."

Those incidents are real, though. OpenAI has disclosed that models undergoing an internal cybersecurity evaluation bypassed controls meant to isolate them from the internet and reached systems belonging to Hugging Face. Google's Gemini accessed protected systems at three real companies during a security test after internet access was unintentionally left open. Australian Prime Minister Anthony Albanese said an OpenAI agent accessed a government health website, calling it a matter of "extreme concern," according to Fox Business.

Taiwan as the test case

The clearest evidence that AI-assisted hacking is already operational comes out of Taiwan. Reporting compiled by The Ins Press lays out a sequence from this past August: the Financial Times reported that Israeli firm Dream identified what it called a near-autonomous AI attack on a government, built from two open-source agent frameworks that ran up to eight agents over four days in early July. The tool mapped 21 Taiwanese government systems, compromised at least 85 accounts, and pulled more than 2,500 personnel records before turning toward Taiwan's nuclear safety agency and several energy companies. Dream said the attack bypassed AI safeguards by falsely labeling itself an "authorized penetration test." Dream has not identified which AI model was used, and Taiwan's Ministry of Digital Affairs confirmed an AI-agent intrusion occurred without naming the model or reporting any disruption to services.

Twelve days after that report, Bloomberg reported that Taiwanese threat-intelligence firm TeamT5 watched state-affiliated Chinese hacking groups more than double their attack volume after offloading reconnaissance and malware development to DeepSeek and other open-source AI models. TeamT5's chief analyst described DeepSeek as "relatively powerful with very low cyber guardrails," according to Bloomberg. Taiwan's National Security Bureau logged 2.4 million intrusion attempts a day against government networks in 2024, rising to 2.63 million a day in 2025.

Two days after the Bloomberg report, on August 26, the U.S. Justice Department seized the domains behind QScan and QTRouter, described as a scanning platform and a proxy routing tool, according to The Ins Press.

The Justice Department has not publicly detailed what role, if any, those domains played in the Taiwan campaign, and no charges tied to the seizure have been announced. Whether U.S. and Taiwanese authorities can name the specific AI model behind the Dream-identified attack, and whether Beijing-linked groups keep scaling up DeepSeek-assisted operations now that the pattern has been documented, remains open.