AI spending and safety face scrutiny as Anthropic prepares IPO

Anthropic’s updated IPO filings showed US$17B in Q3 revenue alongside a US$23.3B net loss and US$895B in compute commitments. Compute costs reached 58% of revenue, underscoring the financial pressure behind rapid growth. The company also disclosed that it may imminently achieve recursive self-improvement, requiring unprecedented safety measures. OpenAI reduced its funding target to US$20B at a US$1.25 trillion valuation amid technology stock selloffs.

Governments are moving toward stronger oversight. Britain allocated £120m to its AI testing lab, while China planned safety standards spanning development and deployment. Democrats, after winning control of Congress, intend to investigate AI companies and pursue tighter safety and data center regulations. Community opposition is also pushing infrastructure developers to offer stronger commitments on resource use, transparency and local costs.

Agent security and reliability remain significant obstacles. An Accenture contractor gave an AI agent access to the Pentagon’s secure network, leading to leaked credentials and sensitive data. CrowdStrike introduced controls to terminate and isolate rogue agents, while Anthropic released a cryptographic verification prototype. Commercial adoption remains uneven, with DoorDash reporting only several hundred orders from third-party AI agents.

Investment nevertheless continues across infrastructure and specialized applications. Fal raised US$1.3B at a US$16B valuation, and Sentra raised US$150m to expand security controls over agent access to corporate data. Salesforce agreed to acquire voice AI platform Cresta for approximately US$2.7B.