On June 22, a man in Poway, California received a call at work from someone claiming to be a police officer, who said his daughter had caused a crash that injured a pregnant woman. The caller knew a private medical detail from her childhood — enough to make the story "check out." Then a woman came on the line, crying, saying she was sorry. He later said it sounded exactly like his daughter's voice, "to a T."
He and his wife withdrew nearly all their savings and handed the cash to a courier in a parking lot, only realizing the deception after reaching his real daughter by phone. The FBI has separately warned that AI-cloned voices, built from just seconds of audio scraped off social media, are now driving a wave of family-emergency scams nationwide.
Security researchers note the goal isn't a perfect voice replica — just enough emotional shock to short-circuit verification before a victim can think clearly.
The cap follows months of Tesla actively pushing employees toward heavier AI use, including internal dashboards that ranked staff by token consumption. Some software engineers were reportedly burning through thousands of dollars a week on model usage.
The most-discussed detail is the carve-out: spending on beta xAI products, including Grok and the Composer coding tool, doesn't count against the limit — while usage of Anthropic, OpenAI, and Google models does. Several outlets have read that as a way to steer employees toward Musk's own AI ecosystem.
The move mirrors caps already imposed at Uber, Meta, and Walmart as token-based billing exposes companies directly to the cost of every prompt.
The discussions are exploratory: Anthropic hasn't decided what the chip would do, how powerful it needs to be, or how it would fit into a server, and the project may not proceed at all. The company recently hired Clive Chan, an early member of OpenAI's own chip team, adding weight to the signal.
Anthropic told reporters that a diversified hardware stack — Google TPUs, Amazon Trainium, and Nvidia GPUs — "will remain pivotal" to its compute strategy regardless of the Samsung talks. Samsung, along with SK Hynix and Micron, was also a strategic investor in Anthropic's $65 billion fundraising round in May.
Isaac 1 rolls on a wheeled base and rises from a crouch to 5'9" when there's work to do. At $7,999 upfront or $449 a month, it undercuts bipedal rivals like 1X's Neo (~$20,000) and Figure or Unitree (from $12,000), largely because legs demand expensive actuators and sensors.
Deliveries begin this fall, but only in California; the rest of the US waits until 2027, with Europe not yet on the map. Autonomy is partial — the robot leans on remote human teleoperation when a task proves too difficult, and Weave hasn't clarified whether in-home footage is used to train the system.
China now hosts roughly 150 companies building humanoid robots, with leaders like Unitree and Agibot iterating quickly on the back of the country's EV supply chain. But researchers describe most current systems as sitting at "level 0–1" autonomy — capable only of scripted or remotely operated tasks — with the software brain still lagging well behind the hardware body.
One analyst called the scaling bet "not a super-crazy idea," but stressed there isn't yet evidence that teleoperation data and egocentric video translate into robots that can function reliably in arbitrary, real-world environments.
The feature sits under Tools → Audio, offering "Listen to this tab" for a full read-aloud or "Listen to document summary" for a brief spoken synopsis, typically under a few minutes. Playback speed runs from 0.5x to 2x, and users can switch voice styles mid-session.
It's rolling out gradually to Google AI Pro and Ultra subscribers, along with select Workspace Enterprise and Education plans — bringing Docs a step closer to NotebookLM-style audio overviews.
The Commerce Department imposed the export-control order three days after Fable 5 and Mythos 5 launched on June 9, following a jailbreak reported by Amazon researchers. Anthropic responded with an updated classifier that blocks the technique in more than 99% of cases and reroutes flagged requests to Claude Opus 4.8.
Fable 5 is now available globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork; Mythos 5 remains limited to vetted US organizations under Project Glasswing. Anthropic also launched a HackerOne channel for researchers to report future jailbreak attempts.