Google's generative AI is hallucinating real people's phone numbers in response to business queries, flooding private individuals with misdirected calls. This is a live, at-scale harm — not a theoretical risk — affecting ordinary users with no recourse. It illustrates that RAG and grounding are insufficient safeguards when models confidently confabulate contact information.
OpenAI launched a personal finance feature for ChatGPT Pro users that lets them securely connect bank accounts and receive AI-driven insights on spending, subscriptions, portfolio performance, and upcoming payments. This moves ChatGPT directly into fintech territory, threatening consumer products like Mint successors, Copilot, and robo-advisory dashboards. The feature is US-only initially, suggesting a careful regulatory rollout strategy.
TechCrunch's coverage adds detail to the OpenAI personal finance launch: users get a structured dashboard view of portfolio performance, spending, subscriptions, and upcoming payments alongside the chat interface. The dual-mode product (dashboard + conversational) is a notable UX choice that hedges against users who distrust pure chat for financial decisions. This is a direct competitor to Monarch Money, Copilot, and any B2C fintech built on 'smart money insights.'
OpenAI acquired Weights.gg, a ~6-person startup that enabled user-generated AI voice clones of celebrities, folding the team into OpenAI without plans for a standalone product. The acqui-hire signals OpenAI is aggressively building voice synthesis depth into its core platform rather than spinning it out. The startup's notoriety for celebrity imitation clones is notable — OpenAI is buying technical and UX expertise in a space with serious consent and legal exposure.
Anthropic is raising another $30B round just three months after a prior $30B raise, pushing its valuation to $900B — surpassing OpenAI's for the first time — on the back of annualized revenue approaching $45B, a 5x increase in roughly a year. This is not a speculative valuation; the revenue multiple, while aggressive, is grounded in real enterprise traction. The speed of successive $30B rounds suggests Anthropic is deploying capital into compute and infrastructure at a rate that would be alarming if revenue weren't tracking.
Cerebras completed a $5.5B IPO with shares doubling on day one, making it the first major AI infrastructure company to successfully go public in 2026. This validates the thesis that purpose-built AI silicon is a durable market, not just an Nvidia adjacency story. A 108% pop signals strong public market appetite for AI infrastructure plays, which will accelerate the IPO pipeline for other infrastructure names.
Databricks has integrated GPT-5.5 into its enterprise agent workflow platform after the model achieved state-of-the-art performance on the OfficeQA Pro benchmark, which tests document and enterprise data question-answering. This partnership puts GPT-5.5 at the center of data engineering, analytics, and enterprise AI agent pipelines for Databricks' large enterprise customer base. It also signals OpenAI's strategy of embedding into data platform partners rather than building its own enterprise data infrastructure.
CMU researchers built a benchmark targeting autonomous exploitation of real V8 engine vulnerabilities, and both Claude Mythos and GPT-5.5 can now produce working browser exploits without human assistance. Claude Mythos leads by a wide margin but costs 12x more than GPT-5.5, creating a meaningful cost-capability tradeoff. This is a significant dual-use inflection point: the same capability that threatens security is a breakthrough for automated offensive security research and defensive tooling.
Richard Socher's new startup raised $650M to build a self-improving AI system — one that can autonomously research and iterate on its own architecture and training. The pitch is that recursive self-improvement is the path to escaping the current paradigm of expensive, human-directed training runs. This remains highly speculative, but the capital commitment signals serious investor belief that the 'AI builds AI' thesis is worth a major bet.
Runway is repositioning from a video generation tool to a world model research lab, betting that generating coherent video is the best training signal for building models that understand physical reality. The strategic claim is that their outsider status — not being tied to search, ads, or existing AI product lines — is an advantage in pursuing this direction. This mirrors the thesis Waymo and DeepMind have pursued: video prediction as a path to generalized world modeling.
That's today's briefing.
Get it in your inbox every morning — free.
Help us improve AI in News
Got a suggestion, bug report, or question?