# AI Brings Back the Worm Era, Lackey Warns _Zane Lackey says AI offense revived 1990s zero days and worms, forcing faster patching and new hardened controls._ **Published:** 2026-11-16 **Source:** https://www.startuphub.ai/ai-news/investors-news/2026/ai-brings-back-the-worm-era-lackey-warns --- Zane Lackey, writing for [a16z](https://www.a16z.news/p/the-dream-of-the-1990s-is-alive-again), argued that Opus 4.5 and 4.6 reset the economics of offense and revived the late 1990s pattern of emergency zero days. He pointed to May 1998, when seven L0pht hackers told the Senate they could take down the Internet in 30 minutes, and to 2002, when remote flaws hit Apache, OpenSSH, OpenSSL and BIND. He said security teams now spend most of 2026 clearing calendars the way they did then. AI Offense AcceleratesDriveropus 4.5 and 4.6 reset economics of offense to machine speedFrom the article 3 mentionsZane Lackey, writing for a16z, argued that Opus 4.5 and 4.6 reset the economics of offense and revived the late 1990s pattern of emergency zero days.outpacesPatching Lags BehindDriveroffense runs at machine speed while patch windows lag behindenablesWorm Era ReturnsOutcomeself propagating internet worms predicted to return at scale soonFrom the articleLackey predicted self-propagating Internet worms will return because AI collapses discovery and exploit time while patch windows lag.stealsAPI Keys TargetedEffectFrom the articleHe said the new worms will steal API keys for inference, which he described as displacing crypto mining, plus ransomware.forcesOverlapping Controls NeededOutcomefully working controls still bypassed requiring overlapping hardened tuning dailyFrom the articleHe added that fully working controls will still get bypassed by zero days and speed, so teams must tune overlapping detection for response at AI speed.Opus CatalystCorea16z author zane lackey cites opus models as offense catalystFrom the articleZane Lackey, writing for a16z, argued that Opus 4.5 and 4.6 reset the economics of offense and revived the late 1990s pattern of emergency zero days.drivesAI Offense AcceleratesDriveropus 4.5 and 4.6 reset economics of offense to machine speedFrom the article 3 mentionsZane Lackey, writing for a16z, argued that Opus 4.5 and 4.6 reset the economics of offense and revived the late 1990s pattern of emergency zero days.Patching Lags BehindDriveroffense runs at machine speed while patch windows lag behind1990s Zero DaysContextseven l0pht hackers warned senate in may 1998 plus 2002 apache flawsFrom the article 2 mentionsZane Lackey, writing for a16z, argued that Opus 4.5 and 4.6 reset the economics of offense and revived the late 1990s pattern of emergency zero days.echoesSlammer Sasser HistoryContextsqlslammer and sasser spread via remote flaws knocking networks offlinepreviewsWorm Era ReturnsOutcomeself propagating internet worms predicted to return at scale soonFrom the articleLackey predicted self-propagating Internet worms will return because AI collapses discovery and exploit time while patch windows lag.stealsAPI Keys TargetedEffectFrom the articleHe said the new worms will steal API keys for inference, which he described as displacing crypto mining, plus ransomware.forcesOverlapping Controls NeededOutcomefully working controls still bypassed requiring overlapping hardened tuning dailyFrom the articleHe added that fully working controls will still get bypassed by zero days and speed, so teams must tune overlapping detection for response at AI speed. His thesis is simple: offense now runs at machine speed and patching does not. Lackey predicted self-propagating Internet worms will return because AI collapses discovery and exploit time while patch windows lag. He recalled SQLSlammer and Sasser, which spread through remote flaws and at times knocked networks offline with scan traffic. He said the new worms will steal API keys for inference, which he described as displacing crypto mining, plus ransomware. He added that fully working controls will still get bypassed by zero days and speed, so teams must tune overlapping detection for response at AI speed. Joel de la Garza, credited as collaborator on the essay, helped frame the prescription for defenders. Lackey urged continuous no-scope red teaming with current attacker tooling, including specialized post-trained models sold by Hinoki Security, plus aggressive internal testing for initial footholds. He pointed to startups building hardened replacements, naming Neo for endpoint, Keycard for agent identity and Socket for open source supply chain, and he recalled OpenBSD as the earlier model. He conceded that prior hardened components struggled commercially. That bet on new identity sits uneasily beside what incumbents claim they already ship. [Okta](https://www.okta.com/blog/product-innovation/okta-mcp-server-use-cases/) says its MCP Server lets agents including [Claude](https://www.startuphub.ai/ai-news/technology/2026/claude-code-s-latest-features), GitHub Copilot and Cursor query and automate an Okta org with natural language. [Okta for AI Agents](https://www.okta.com/blog/ai/shadow-ai-securing-personal-ai-agents/) is described as working with CrowdStrike Falcon to find untrusted agents on endpoints and tie them to workforce identities for governance. Lackey described Keycard as the new Okta for agents, while Okta described extending its own platform to govern those agents. Both cannot be the default control. Lackey also revived the Windows XP problem, citing the SANS guide Surviving the First Day, when a fresh machine was compromised before updates finished. He said enterprises resisted auto updates over breakage risk and unmaintained software, but AI offense will force patching orders of magnitude faster. How short must that window get before auto updates break less than the worms do? --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.