Google flags urgency as AI reshapes cyber threats

Google Threat Intelligence is warning that artificial intelligence is lowering the technical threshold for cyber attackers, helping threat groups scale activity, speed up intrusions and automate parts of their operations.

"Right now, there is a very big sense of urgency to patch very quickly," said Sandra Joyce, VP, Google Threat Intelligence, Google.

Joyce said Google is seeing attackers use AI in three main ways: scale, speed and sophistication. Fully agentic, mass-scale AI attacks have not yet been widely observed in the wild, but she said the underlying capability is developing quickly enough to change how organisations assess risk.

Agentic attacks

Attackers are experimenting with open source tools that bring large numbers of AI-enabled capabilities into one place. Joyce cited HexStrike MCP, a tool designed for red teams that Google has also observed in threat actor use.

The speed of intrusion activity is also changing. Attackers previously had to pause during operations to research next steps, seek help, or work through privilege escalation and network enumeration over days or weeks. Some are now turning to AI systems for guidance during operations.

One China-based actor, Joyce said, attempted to disguise malicious activity by framing it as a capture-the-flag exercise when asking Gemini for advice. Google has since updated classifiers to prevent that behaviour on Gemini.

"I think we're going to see it very, very soon," said Joyce.

The immediate concern is not that every attacker can already press a button and launch a fully automated attack. It is that the skills needed to scan for vulnerabilities and write code are becoming more accessible. Joyce said some capability still requires technical understanding, but the direction is clear.

"For organisations that have not patched and have not done very good patch management, they're going to have a real problem as these tools become better and better at scanning for vulnerabilities, which many of them can already do," said Joyce.

Dark web

Google is also using AI inside its own threat intelligence operations. Joyce said Google launched agentic capabilities for dark web analysis at RSA, using Gemini to improve how underground chatter is interpreted.

Traditional search-and-match methods produced high false positives, particularly because dark web conversations rely on slang, dialect and language specific to underground communities. Google trained Gemini to process those patterns across languages, including Russian and Chinese, as well as the way threat actors communicate in criminal forums.

"We're looking now at a 98% accuracy rate versus a 90% false positive rate," said Joyce.

Inside Google Threat Intelligence, analysts are also moving through an AI transformation. The initial focus was productivity, including faster malware reversing and workflow automation. That work has now shifted towards building agents and, eventually, managing groups of agents that can help interpret the threat landscape.

Headcount may not change dramatically, Joyce said, but the work will. Analysts will be expected to use AI to handle larger data volumes, identify relevant signals faster and manage automated systems that support investigations.

Wiz integration

Google has begun integrating its threat intelligence into Wiz following the close of its acquisition of the cloud security company.

The goal is to bring more real-time intelligence from incident response and Google Threat Intelligence into the Wiz platform. That would allow customers to assess cloud risk against current threat actor behaviour, rather than older techniques.

Wiz customers are expected to see a more enriched experience over time, including red teaming informed by recent attacker activity and remediation supported by Mandiant expertise.

"We are already very deep into integrating our capabilities together," said Joyce.

One priority is increasing the speed at which incident response findings reach Wiz users. That requires engineering work, but the aim is to ensure intelligence from current intrusions can inform red teaming, defence and remediation more quickly.

"We're going to Red Team you, not with some technique that's six months old. We're going to Red Team you with what this threat actor was doing last week or earlier this week," said Joyce.

Disruption work

Google Threat Intelligence has also expanded its disruption work, pairing technical action with legal measures and industry partnerships to take down malicious infrastructure.

Joyce cited action against IPidea, a residential proxy network used by more than 500 threat actors, including actors linked to Russia, China, North Korea and Iran, as well as criminal groups. Threat actors used the network to disguise activity by making traffic appear to come from ordinary consumer devices.

Google took legal action against companies selling the service, carried out technical takedowns, seized infrastructure and sinkholed parts of the network. Joyce said partners including Okta observed a 90% reduction in IPidea exit nodes.

Another disruption targeted a China-based espionage campaign aimed at telecommunications providers and governments in 42 countries, particularly across Asia. Joyce said the actor had used Google Sheets for command and control. Google connected the malware and victimology through Mandiant threat defence work.

Cyber criminals' behaviour has also changed, Joyce said, with fewer apparent limits on the sectors they are willing to target.

"We used to see threat actors say things in the underground like, 'We're not going to go after hospitals.' We don't see that anymore," said Joyce.