Today in AI — 29 April 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
OpenAI's CFO is warning that slowing growth can't support $600 billion in compute commitments, and the company just missed its own revenue targets. The uncomfortable question this week: what happens when the capital bets outpace the revenue that's supposed to justify them?
The billion-dollar bets
Google's $40 billion Anthropic commitment and Cognition's $25 billion ask arrive the same week OpenAI missed its own revenue and user growth targets. The capital is flowing faster than the revenue it's chasing.
- OpenAI misses revenue and user growth targets as CFO clashes with Altman over $600 billion in compute commitments — Fortune
- Google to invest up to $40 billion in Anthropic in cash and compute — CNBC
- Cognition AI in funding talks at $25 billion valuation as Devin gains enterprise traction — SiliconANGLE
Model competition fragments
DeepSeek's V4 keeps compressing the gap between frontier and open-source with 1.6 trillion parameters and MIT-licensed weights. Microsoft launched three proprietary models competing directly with its own partner OpenAI. Cohere merged with Aleph Alpha to build a $20 billion sovereign alternative for regulated industries. The model layer is splintering along geographic and regulatory lines, not consolidating.
- DeepSeek releases V4 open-source model with 1.6 trillion parameters and million-token context — CNBC
- Microsoft launches three in-house AI models in direct challenge to OpenAI and Google — VentureBeat
- Cohere and Aleph Alpha merge to form $20 billion sovereign AI powerhouse — TechCrunch
Reliability scales slower than capability
An ICLR 2026 paper found that training models for stronger reasoning proportionally increases hallucination rates. That finding lands at the very conference where GPTZero identified 50+ hallucinations in accepted papers. OpenAI's apology to Tumbler Ridge after failing to flag a mass shooting suspect's ChatGPT activity is a stark reminder that safety obligations scale with deployment, not intentions.
- Making AI models smarter also makes them hallucinate more, ICLR paper finds — OpenReview
- GPTZero uncovers 50+ AI-generated hallucinations in ICLR 2026 accepted papers — GPTZero
- Sam Altman apologizes to Tumbler Ridge for not alerting police about mass shooting suspect's ChatGPT use — TechCrunch
Where agents earn their keep
The clearest wins are specific: Avoca hit $1 billion answering missed calls for plumbers and HVAC companies, Anthropic shipped persistent memory for Claude Managed Agents, and Claude Mythos found 271 Firefox vulnerabilities that existing tools had missed.
- The AI voice agent for plumbers just hit a $1 billion valuation — Fortune
- Anthropic adds persistent memory to Claude Managed Agents in public beta — SD Times
- Mozilla patches 271 Firefox vulnerabilities found by Anthropic's Claude Mythos — SecurityWeek
The money flowing into AI this week could fund a small nation's annual budget; whether it pays off depends less on the models and more on whether builders can close the gap between demo performance and production reliability.