In 2025, AI didn't create new hackers: It made existing ones faster, cheaper and harder to stop.

Why it matters: Even if models stop improving — which is unlikely — the barrier for who can launch a cyberattack has already fallen dramatically.

The big picture: 2025 marked the first time AI crossed from experimentation into sustained use inside real-world cyber operations — as Anthropic documented in its report on Chinese state hackers using Claude to automate 80%-90% of a cyber-enabled espionage campaign.

• That disclosure triggered a congressional hearing last week focused on how lawmakers should intervene, if at all.

Between the lines: Hackers and fraudsters began deploying AI at scale to compress timelines and reduce human effort across the attack chain.

• In the case of North Korean IT worker fraud, that involved using AI to convincingly assume stolen personas during remote job interviews.
• Russian state operators used AI tools to assist in writing and modifying malware code, according to Google researchers.
• Cybercriminal groups tapped deepfake voice and video tools to impersonate C-suite executives and family members in financial fraud schemes.
• China-based influence operators used generative AI to build adaptive disinformation campaigns capable of mining social media profiles and generating more believable content at scale.
• Security researchers also found evidence that attackers are using AI tools to accelerate vulnerability discovery — shrinking the window between a flaw's identification and its exploitation.

Zoom in: Cynthia Kaiser, a former top FBI cyber official who now leads Halcyon's ransomware research center, told Axios that her team has uncovered ransomware gangs using AI tools in their attack chains.

• In July, researchers at Picus Security said they'd uncovered evidence of ransomware gangs using AI chatbots to automate negotiations with victim organizations.

Yes, but: All of these instances still require a human operator — for now.

The intrigue: Defenders are racing to use the same tools against attackers and are testing whether AI can help detect, analyze and respond to threats in real time.

• A Stanford paper published this month detailed how an AI agent found bugs in one of the university's networks with no assistance — highlighting the power of AI agents for defense.
• And cybersecurity dealmaking and investments lit up in 2025 as defenders raced to outsmart adversarial hackers.
• "The question isn't whether AI will transform the threat landscape, but whether defenders can adapt fast enough," Josh Kamdjou, CEO and co-founder of Sublime Security, told Axios. "The good news is that AI works both ways."

What to watch: Whether companies actually pick up and use these AI-driven defenses quickly enough.

The top U.S. cyber agency is heading into 2026 on rocky footing after lawmakers left town without approving President Trump's pick to lead the agency and a startling story involving a failed polygraph test landed over the weekend.

Why it matters: The Cybersecurity and Infrastructure Security Agency helps man the digital barricades for U.S. government networks and critical infrastructure, including ports, water systems and the electric grid.

• Its slow start at the beginning of the second Trump administration could weaken the country's cyber defenses at a moment when AI is accelerating malicious hackers.

Driving the news: Politico reported Sunday that at least six career staffers at CISA were placed on administrative leave over the summer after organizing a polygraph test for the agency's acting director, Madhu Gottumukkala. He later failed the test.

• According to the report, Gottumukkala, who is also CISA's deputy director, took the test to access a sensitive intelligence program. The previous deputy director did not see this program, per the report.
• In a statement, DHS spokesperson Tricia McLaughlin told Axios that Gottumukkala took an "unsanctioned polygraph test" and that the staff who coordinated the test were "misleading incoming CISA leadership."
• "The employees in question were placed on administrative leave, pending conclusion of an investigation," she added. "We expect and require the highest standards of performance from our employees and hold them directly accountable to uphold all policies and procedures."

Between the lines: All of this is happening after the agency lost at least a third of its staff this year to buyouts and layoffs — and after Congress failed to vote on the president's nomination of Sean Plankey to run CISA before lawmakers left for the holidays.

• "We're a sinking ship. We're like the Titanic," a current CISA official told Politico.

What to watch: The White House will need to resubmit Plankey's nomination in the new year, and it remains unclear whether the administration plans to continue backing him.

• The Trump administration is also working on a new cyber strategy, expected as soon as next month. How prominently CISA features in a subsequent detailed action plan may offer early clues about the agency's future.

ServiceNow is shopping in two fast-growing areas of cybersecurity — securing agentic AI and threat exposure management.

Why it matters: The IT service management platform's wish list hints at which cyber subsectors will be the hottest for dealmaking in 2026.

Driving the news: ServiceNow said this morning it would acquire San Francisco-based cyber exposure management company Armis for $7.75 billion cash.

• Earlier this month, ServiceNow said it would acquire startup Veza for more than $1 billion to gain better visibility into human, machine and AI agent identities.

Between the lines: The need to secure AI agents and nonhuman identities is pretty self-evident, but the Armis deal left some scratching their heads.

Yes, but: While Armis is considered an IoT and operation technology security company, its platform also allows users to monitor devices and understand their vulnerabilities.

What we're hearing: The Armis deal also likely serves as another entry point for ServiceNow to tackle the operational technology sector, helping grids and factories with their IT and security, a VC tells Axios Pro.

Catch up quick: Santa Clara, California-based ServiceNow inked at least six acquisitions this year before the Armis move — including MoveWorks, Logik.ai, Quality 360 and Cuein.

The bottom line: Expect more cyber deals in 2026, with an emphasis on agentic AI security and threat exposure management assets.

💡 If you need smart, quick intel on dealmaking in the enterprise software industry for your job, get Axios Pro Deals.

@ D.C.

🪖 Lt. Gen. Joshua Rudd appears to be the Trump administration's new pick to lead U.S. Cyber Command and the National Security Agency. (Politico)

🎓 CISA is now participating in the U.S. government's CyberCorps scholarship program, which will make 100 CISA internships available to students. (Nextgov)

@ Industry

💰 Kevin Mandia, founder of cyber firm Mandiant, has started a new company to help defenders detect vulnerabilities in their networks ahead of the expected onslaught of AI hacks. (Wall Street Journal)

👀 The founder of spyware maker NSO Group has been trying to reinvent himself as he builds a new critical infrastructure security startup. (Bloomberg)

@ Hackers and hacks

⚠️ Cisco warned that China-backed hackers are exploiting a new zero-day in Cisco AsyncOS software. (TechCrunch)

📸 Flock Safety — the operator of controversial license plate readers taking over U.S. cities — left its livestreams and administrator control panels for at least 60 of its AI-enabled cameras exposed on the open internet. (404 Media)

🥸 The FBI says nefarious actors have been using AI voice-cloning tools to impersonate U.S. government officials since 2023. (CyberScoop)

📉 Sure, AI is good, but it's not amazing. Just check out what it did when it ran a vending machine in the WSJ's newsroom for three weeks.