AI
Every Attack Was a Warning Shot, Not an Anomaly
Perlroth covered what she calls “a decade of firsts” at the New York Times: the first time Google announced China hacked them, the first attack through an HVAC system, the first time cyber weapons sabotaged the physical world. None were truly first. Stuxnet was part of Operation Olympic Games. Shamoon borrowed its wiper capability from that same program. NotPetya iterated on the MeDoc accounting-software vector. SolarWinds refined that again.
“These weren’t black swans, they were warning shots.”
Every successful attack emboldened the next actor. The industry treated each as an edge case, moved on after a news cycle, and expanded the attack surface in the meantime.
The Human Toll Reporters and Defenders Kept Underestimating
When ransomware hit University of Vermont Medical Center, nurses told Perlroth: “We can’t give our chemo patients their cancer treatments. We can’t tell them if their tumor has grown.” One nurse compared the trauma to working the burn unit after the Boston Marathon bombing. A 78-year-old woman in Germany died in transit after a ransomware attack took down the nearest hospital.
The press covered these incidents for a news cycle and moved on. Meanwhile, attacks kept escalating. Ransomware spread from individual PCs to enterprises to supply chains, then to Colonial Pipeline and United Healthcare, each iteration more damaging.
Disinformation as a Weapon Against Businesses and Officials
The CEO of Rio Tinto told Perlroth that Russian bots hit Serbian social media the day his company cut the ribbon on a lithium mine projecting one million EV batteries a year. Conspiracy theories spread about uranium and sulfuric acid. Tens of thousands protested. The Serbian prime minister shut the project down. See the Rio Tinto Serbia mine disinformation campaign (28:56) for the full sequence.
“I was prepared for a $50 million ransomware attack. I was not prepared for a $2.4 billion disinformation attack.”
By 2025, that playbook had gone domestic: RSA panelists were pulled from Perlroth’s sessions because employers feared federal contract loss if speakers mentioned Chris Krebs.
Volt Typhoon and the China Prepositioning Strategy
Nick Lawler manages utilities for about 15,000 people in Littleton, Massachusetts. CISA and the FBI called him to say Volt Typhoon had been inside his systems. He verified the agents, told them to show up Monday at 10 a.m., and together they rooted it out. See Volt Typhoon in Littleton Electric systems (37:31).
Volt Typhoon gets in through compromised home routers, default credentials, misconfigurations. It converts them into botnets and parks inside critical infrastructure. The 1999 PLA doctrine “unrestricted warfare” spells out the goal: disable power, water, and communications to diminish U.S. political will to defend Taiwan.
AI Widens the Attack Surface Before Defenders Are Ready
Microsoft Copilot was broken in June 2025. Ransomware groups now automate the kill chain: scanning every CVE, finding critical assets, running AI chatbots through ransom negotiations. A Veracode report found LLMs perform poorly at generating secure code. Ken Thompson’s 1983 warning applies to vibe coding: you cannot trust code you did not write yourself.
McKinsey found 75% of companies have adopted AI but only a quarter moved past pilots. Once AI is embedded in defense decisions, the cost of failure multiplies. Perlroth: secure raw data before it reaches models, red team relentlessly, and find the courage to say so publicly.
Notable Quotes
These weren’t black swans, they were warning shots. Nicole Perlroth · ▶ 23:42
I was prepared for a $50 million ransomware attack. I was not prepared for a $2.4 billion disinformation attack. Nicole Perlroth · ▶ 31:16
we’re not in the cyber security business anymore. We are in the saving civilization business. Nicole Perlroth · ▶ 51:45
Key Takeaways
- Each decade of breaches from Stuxnet to SolarWinds was a preview, not an anomaly; treating them as edge cases let the threat metastasize.
- Volt Typhoon is prepositioning inside U.S. critical infrastructure via home-router botnets, staging for a Taiwan contingency, not espionage.
- AI currently favors offense: ransomware groups automate the kill chain while LLMs produce insecure code, and the window to fix this is closing fast.
About the Speaker
Nicole Perlroth
Founding Partner at Silver Buckshot Ventures. Cybersecurity reporter and author, formerly with New York Times.