Google Stops AI Exploit, Nadella Testifies, OpenAI's New $4B Unit

Summary of Google Stops AI Exploit, Nadella Testifies, OpenAI's New $4B Unit

by Lex Fridman Podcast Fan

16m•May 11, 2026
Listen to Episode

Overview of Google Stops AI Exploit, Nadella Testifies, OpenAI's New $4B Unit

This episode covers a fast-moving slate of AI and tech news: the EU opening direct talks with OpenAI and Anthropic about AI Act enforcement, OpenAI creating a dedicated $4 billion enterprise-focused unit, Satya Nadella’s testimony in the Elon Musk v. OpenAI lawsuit, and Google’s disclosure that it stopped what it believes was the first AI-assisted zero-day exploit seen in the wild. The discussion also touches on Baiju Bhatt’s new space startup raising $275 million to build rockets and orbital data centers, positioned as a future competitor to SpaceX.

Key Stories

EU regulators talk directly with OpenAI and Anthropic

  • The European Commission has begun direct discussions with OpenAI and Anthropic about how the EU AI Act applies to frontier models, routers, and AI chat systems.
  • The host frames this as pragmatic, since the biggest real-world AI usage is concentrated among a few major labs.
  • At the same time, critics argue this could become a form of regulatory capture, since the largest labs may end up shaping rules that smaller labs and open-source teams must also follow.

Baiju Bhatt raises $275 million for orbital compute ambitions

  • Robinhood co-founder Baiju Bhatt reportedly raised a $275 million Series B at a $2 billion post-money valuation for a space venture aimed at launching rockets and placing data centers in orbit by 2028.
  • The company’s thesis is that demand for compute will outstrip terrestrial launch capacity, making orbital infrastructure commercially attractive.
  • The episode notes the enormous technical and capital challenge:
    • building rockets and engines,
    • launching heavy payloads,
    • and fitting nearly data-center-scale GPU compute into a satellite/rocket system.
  • The idea is presented as both ambitious and highly speculative, especially given SpaceX’s lead and the difficulty of competing with established launch providers.

OpenAI creates a dedicated enterprise unit

  • OpenAI is forming a new $4 billion enterprise division focused on corporate AI development.
  • The unit will handle:
    • enterprise sales,
    • integrations,
    • professional services,
    • and deployment engineering for large customers.
  • The move suggests ChatGPT Enterprise is no longer just a side offering; OpenAI wants a real product and go-to-market organization for the enterprise market.
  • The episode emphasizes that enterprise contracts are:
    • stickier,
    • longer-term,
    • and higher margin than consumer subscriptions.
  • This also increases direct competition with Microsoft, Salesforce, ServiceNow, Google, and Anthropic in the corporate AI market.

Satya Nadella testifies in Musk v. OpenAI

  • Satya Nadella took the witness stand in Elon Musk’s lawsuit against OpenAI, now in week three of the trial.
  • Musk is seeking massive damages and wants Sam Altman and Greg Brockman removed, arguing that OpenAI breached its original mission and structure.
  • New disclosures from Microsoft’s side revealed:
    • Microsoft has recognized about $9.5 billion in revenue tied to OpenAI over the relationship’s lifetime.
    • Microsoft’s total commitments and Azure compute support amount to roughly $13 billion.
  • The episode highlights how quickly Microsoft may have effectively recouped much of its investment through OpenAI-related revenue.
  • The judge reportedly told Musk’s legal team they were “in mud” on their legal theory, which the host interprets as a setback for Musk’s case.

Google says it stopped the first AI-built zero-day exploit in the wild

  • Google Threat Intelligence Group says it intercepted what it believes was the first AI-generated zero-day exploit observed in the wild.
  • The exploit targeted two-factor authentication on an unnamed open-source web-based system administration tool.
  • Google says the attackers were preparing a “mass exploitation event,” and that it disrupted the campaign before it could spread.
  • The odd clue that exposed the AI involvement:
    • a hallucinated CVSS score,
    • and code that looked overly structured, like AI-generated malware with tutorial-style comments and formatting.
  • Google said it does not believe Gemini was used; the host suggests the exploit more likely resembled code produced by another LLM.
  • The key takeaway is that attackers are increasingly targeting:
    • autonomous AI agents,
    • third-party data connectors,
    • browser tools,
    • MCP-style integrations,
    • and other production AI surfaces.

Main Takeaways

  • AI regulation is lagging behind frontier-model reality.

    • The EU is trying to adapt older rules to much newer model capabilities.
    • The biggest labs are being pulled into shaping the interpretation of the rules.

  • Enterprise AI is becoming the main battleground.

    • OpenAI is moving from being mostly a model provider to building direct enterprise sales muscle.
    • This puts it in more direct conflict with Microsoft and other enterprise software giants.

  • The OpenAI–Microsoft relationship is not frictionless.

    • Nadella’s testimony and the revenue figures show the relationship is commercially deep, but strategically complicated.

  • AI is now part of offensive cybercrime tooling.

    • The Google case suggests attackers are already using LLMs to accelerate exploit development.
    • Detection may increasingly depend on spotting AI-generated “tells,” not just malicious behavior.

  • Connector and agent security is now a first-class issue.

    • As more AI tools connect to apps, APIs, and internal systems, the attack surface expands quickly.

Practical Implications for AI Builders and Teams

Security recommendations

  • Re-audit every third-party connector and tool call used by AI agents.
  • Review trust boundaries between models, tools, and external systems.
  • Watch for prompt injection paths and unauthorized data access.
  • Log and monitor all tool invocations in production.
  • Treat agent and connector security as a core product requirement, not an afterthought.

Business strategy implications

  • If you’re building enterprise AI, expect rising competition in:
    • deployment,
    • sales,
    • compliance,
    • and integration services.
  • If you’re building on AI infrastructure, assume regulation and security scrutiny will intensify.

Notable Quote / Idea

  • The most important warning in the episode is that “adversaries increasingly target the integrated components that grant AI systems their utility” — meaning the tools, connectors, and autonomous actions that make AI useful are also becoming the easiest targets.