The evening briefing.
Today across AI and tech: Geopolitical concerns surface with Anthropic, Waymo recalls robotaxis, and Amazon plans to sell its AI chips.
AI Geopolitics & Security. A significant development saw US officials intervene in a partnership between Anthropic and SK Telecom due to alleged ties between the South Korean conglomerate and China, raising concerns about access to frontier AI models. Meanwhile, Google Deepmind is treating its own AI agents as potential insider threats, implementing an "AI Control Roadmap" with security measures tied to agent capabilities. Separately, experts are warning that AI toy apps for children are tracking users and collecting personal data, highlighting broader privacy risks.
AI Infrastructure & Investment. The Federal Energy Regulatory Commission (FERC) is pushing to fast-track AI data centers onto the power grid by updating interconnection rules for massive power users. This comes as the AI inference startup Baseten is reportedly raising $1.5 billion at a $13 billion valuation, signaling continued strong investment in AI infrastructure. Amazon is also looking to directly challenge Nvidia by selling its own AI chips, a move CEO Andy Jassy believes represents a $50 billion opportunity.
Autonomous System Challenges. Waymo has issued a recall for approximately 3,900 robotaxis after some vehicles drove into closed freeway construction zones, prompting a software fix. In a related setback, Rivian is facing a class-action lawsuit from owners alleging false promises regarding the self-driving capabilities of its early vehicles. These incidents underscore the ongoing technical and legal hurdles in deploying autonomous technologies safely and reliably.
AI Development & Policy. OpenAI is strategically bolstering its team ahead of a potential IPO, bringing on key talent like Transformer co-inventor Noam Shazeer and former Trump AI policy official Dean Ball. The company also claims its ChatGPT health upgrade, powered by GPT-5.5 Instant, now outperforms doctor-written answers in accuracy, clarity, and completeness. On the policy front, Senator Bernie Sanders has introduced a new AI bill that proposes paying Americans $1,000 a year, arguing AI is a public resource built on collective human genius.
Alleged China ties at SK Telecom triggered Anthropic crisis, alarmed US officials
US officials intervened in a partnership between Anthropic and SK Telecom, citing alleged ties between the South Korean conglomerate and China. This intervention raised concerns about access to Anthropic's AI model Claude Mythos through the Project Glasswing partner program.
Waymo recalls 3,900 robotaxis after entering freeway construction zones
Waymo has issued a recall for approximately 3,900 robotaxis after some vehicles drove into closed construction zones. The company implemented a software fix to address the underlying issues.
Longsys's SIM-card-sized 8TB PCIe 5.0 SSD hits 11GB/s, likely hoarded by AI firms
Longsys has unveiled a new mSSD that delivers double the maximum capacity and significantly upgraded read/write speeds, reaching 11GB/s. This SIM-card-sized 8TB PCIe 5.0 SSD is expected to be in high demand by AI firms due to its performance and compact size.
New unpatchable exploit targets Apple devices with A12 and A13 chips
Researchers at Paradigm Shift have detailed 'usbliter8,' a new unpatchable iPhone BootROM vulnerability affecting Apple's A12 and A13 chips. This exploit enables arbitrary code execution on affected devices.
AI inference startup Baseten reportedly raising $1.5B at $13B valuation
AI inference startup Baseten is reportedly nearing a $1.5 billion funding round, which would value the company at $13 billion. This mega-round comes just months after its previous funding, indicating a booming "inference gold rush."
Federal regulator wants to fast-track AI data centers onto the power grid
The Federal Energy Regulatory Commission (FERC) is urging grid operators to update rules for connecting massive power users, aiming to fast-track AI data centers onto the power grid. This initiative seeks to address grid stress and improve affordability for these large-load facilities.
Experts warn AI toy apps for kids are tracking users and collecting personal data
A Cybernews study found that applications tied to 10 different AI toys requested permissions categorized as 'dangerous,' raising concerns about user tracking and personal data collection. Experts warn about the privacy implications for children using these apps.
Snap spins off AI video team into new company, Dotmo, due to costs
Snap is spinning off its internal AI video unit into a new company called Dotmo, citing high costs associated with the development. Current Snap staff will transition to Dotmo to focus on AI video development independently.
OpenAI bringing on big guns in lead-up to its IPO
OpenAI is strengthening its team ahead of a potential IPO, recruiting Transformer co-inventor Noam Shazeer from Google DeepMind and former Trump AI policy official Dean Ball. These strategic hires aim to bolster the company's research and policy expertise.
Bernie Sanders’ new AI bill would pay Americans $1,000 a year
Senator Bernie Sanders has introduced a new AI bill proposing to pay Americans $1,000 a year, arguing that AI is a public resource. The bill makes the case that AI was built using the collective genius of humanity.
ChatGPT's new health upgrade beats doctor-written answers, OpenAI says
OpenAI claims its ChatGPT health upgrade, powered by GPT-5.5 Instant, now surpasses doctor-written answers in accuracy, clarity, and completeness. The company reports a 71 percent drop in the error rate for health-related statements.
Amazon hopes to challenge Nvidia more directly by selling its AI chips
AWS is reportedly in talks to sell its custom AI chips to other data centers, aiming to directly compete with Nvidia. CEO Andy Jassy has indicated this venture could represent a $50 billion opportunity for Amazon.
Midjourney pivots from AI image generation to body scanning medical spa
Midjourney is pivoting its business from AI image generation to a medical spa offering body scanning and 'golden light' treatments. The underlying technology for this new venture is reportedly borrowed from an undisclosed partner.
Google Deepmind treats its own AI agents like rogue employees with office keys
Google Deepmind is adopting an "AI Control Roadmap" that treats its own AI agents as potential insider threats, implementing security measures tied to their capabilities. An analysis of coding tasks revealed that most problems stemmed from overzealous agents rather than malicious intent.