This has been a dizzying few days for artificial intelligence, even by the industry's usual standards. Between a historic market shakeout and two separate demonstrations of AI's growing offensive power in security research, the past 72 hours offered a vivid preview of where the technology — and the anxieties around it — are heading.

Wall Street's AI Trade Gets Tested

The week's biggest shock came from the markets, not the labs. An AI-driven stock sell-off wiped roughly $2.5 trillion in market value off Wall Street, easing only after hitting Asian and European markets hard. The trigger was political: technology stocks rebounded Friday after plunging Thursday as traders assessed the implication of a Democratic sweep in the US midterm elections on artificial intelligence. Chipmakers and "neoclouds" bore the brunt — Nvidia and SpaceX each rose after having plunged roughly 9% the day before, with Nvidia suffering its worst day since January 2025. Analysts framed the volatility as a referendum on the sustainability of the AI infrastructure boom, with growing uncertainty among investors over what the midterm results mean for the AI trade, amid concern that Democratic victories could stymie an investment boom that has powered both the economy and markets.

OpenAI's Model Finds Hundreds of Ways to Hijack Linux

While traders fretted over spending, OpenAI published research showing just how far automated vulnerability-hunting has come. OpenAI revealed that its latest automated pipeline discovered 948 Linux kernel vulnerabilities that could enable attackers to obtain root access. The numbers behind that figure are startling: in a mid-October 2026 research run, the pipeline discovered 2,308 candidate Linux kernel vulnerabilities, 1,724 of which were successfully reproduced, producing 948 working local privilege escalation exploits. Perhaps most notable, the agents used at least 10 previously unknown Linux exploitation techniques, and the exploits worked even against hardened systems — tested against a standard desktop configuration with security protections such as KASLR, SMEP, and SMAP enabled. OpenAI says it's disclosing the flaws responsibly through its Patch the Planet initiative, but the episode underscores how offensive AI capability is now outpacing defenders' patch cycles.

DeepMind Teaches Agents to Pick Old Locks

Google DeepMind delivered its own security bombshell a day later. Working with outside cryptanalysts, the team identified two weaknesses in 1024-bit RSA schemes: one that makes signature forgery approximately 1,600 times more efficient and another that cuts the average cost of factoring keys in large batches by about 30 times. Crucially, the research targets 1024-bit RSA, which has largely been abandoned for stronger keys like RSA-2048, and the company emphasized that modern standards securing banking and messaging are unaffected. Still, the real headline is speed: theoretical discoveries in cryptanalysis that traditionally take years were compressed by AI agents into weeks.

Taken together, these stories paint a picture of an industry whose capabilities are sprinting ahead of both markets' confidence and security norms' ability to adapt.

Sources

AI MarketsArtificial IntelligenceCybersecurityDaily AI UpdateDeepMindOpenAI