The Board Room

Top line: rent the model and own the layers around it, because running, checking, containing and financing AI is now where the money and the risk sit. This week's one act-now item is your executives' exposure to zero-click spyware delivered through ordinary ads, and your CISO should start hardening their devices now. Citizen Lab found that the Israeli firm Discover-Tech used targeted ads to infect devices the moment an ad rendered, with no click needed. CyberScoop and Techpresso report the targets included journalists, NATO, the European Commission, NVIDIA and Arm. The ads ran on more than 120,000 websites. Your executives' phones are exposed through ordinary browsing, and user training cannot stop an attack that needs no click.

Anthropic told investors its Q3 revenue reached $16.8B, up about 45% on Q2, after Q2 grew roughly 140%. The Information reports investors now see an IPO valuation nearer $1.5T than the $2T bankers had floated. The Information also reports that Microsoft and Meta are cutting their internal use of Claude. Buyers who can credibly switch vendors now hold the strongest negotiating position in frontier-model contracts so far.

AINews reports that Modal crossed $1B in annualized run rate. Its agent sandboxes alone bring in $300M. Together, Fireworks and Baseten each report between $0.6B and $1.2B. AWS called a one-prompt credential theft in its AgentCore agent platform 'documented and expected.' Both value and risk are settling in the layers that run, check and contain models, and platform vendors are leaving containment to you.

Bloomberg reports that TSMC beat expectations and raised its outlook, yet a chip index lost $1.6T over six sessions. CoreWeave's debt yields went above 15%. Morning Brew reports that hot September inflation data raised the odds of a Fed hike at the Oct 27-28 meeting from 22% to 35%. Goldman attributes nearly half a point of inflation to the AI buildout. Front-loaded AI commitments financed at today's rates carry more risk than demand signals alone suggest.

CyberScoop reports that Senators Maria Cantwell and Mike Rounds asked the administration to expand Commerce's CAISSI center. They want measurable AI security standards and independent third-party evaluators. AIScoop reports that Army Cyber Command is giving AI agents more network operations roles, while House Democrats ask the FAA to pause its SMART tool. Independent evidence that your AI has been tested is becoming a sales asset before any law requires it.