AI news5
Federal judge blocks Pentagon's blacklisting of Anthropic as unlawful
US District Judge Rita Lin ruled on 28 August that the Pentagon's designation of Anthropic as a "supply chain risk" was unlawful, writing in a 59-page order that "the empty invocation of national security is not a blank check to punish and retaliate against government critics." Defense Secretary Hegseth's designation followed Anthropic's refusal to let the military use Claude for surveillance or autonomous weapons, and it blocked the company from certain military contracts. Anthropic sued in March, arguing the designation violated its First Amendment rights and denied it due process.
cnbc.com ↗Why it matters
this is the first time a court has struck down a government agency's AI-safety-motivated retaliation against a lab, a legal precedent that a company's public safety commitments can't be punished as a national-security risk.
Andreessen Horowitz raises $1.1B fund to bankroll AI's physical infrastructure
a16z announced on 28 August it has raised $1.1 billion for a new "Machine Age Fund" dedicated to AI chips, memory, networking, storage, data centres and robotics rather than software. It's the firm's first dedicated hardware-infrastructure vehicle, with general partner Martin Casado citing capacity constraints across the entire hardware supply chain as the reason for the shift.
techcrunch.com ↗Why it matters
one of software investing's most prominent names betting big on physical AI infrastructure signals investors now see the hardware bottleneck, not model quality, as the constraint worth funding.
Hugging Face starts selling a $399 open-source robot you can retrain yourself
Hugging Face and its Pollen Robotics unit opened pre-orders on 27 August for Microduck, a 25cm biped robot with 15 motors, a camera and a LiDAR sensor that ships before Christmas. The robot's SDK, simulation environment and reinforcement-learning training stack are open-sourced under Apache 2.0, so owners can inspect and retrain the seven policies it ships with.
techcrunch.com ↗Why it matters
a sub-$400, fully open-source robot with a real RL training pipeline lowers the bar for hobbyists and researchers to experiment with physical AI, the same democratisation pattern that made open-weight language models mainstream.
Meta closes a loophole that let its smart glasses record with the privacy light hidden
Meta rolled out an update on 28 August that stops its Ray-Ban smart glasses from recording if the capture LED is covered mid-recording, closing a gap where users could start filming first and hide the light afterward. Meta says several third-party services that sold LED-tampering fixes have shut down since it began tightening the safeguard.
9to5google.com ↗Why it matters
covert recording was the biggest trust problem with camera-equipped smart glasses, and closing it after launch rather than before shows how reactively these privacy controls are still being built.
UK actors demand a legal right to own their own voice against AI cloning
More than 80 performers including Nicola Coughlan, Matt Lucas and Hugh Bonneville backed a campaign launched 28 August calling on the UK government to legislate a statutory right for people to own their voice and control AI cloning of it. The "Save Our Voices Now" campaign points to Denmark's recent reforms giving individuals ownership over their face, body and voice as a model, and a public petition has also launched.
variety.com ↗Why it matters
this turns voice-cloning regulation from a hypothetical harm into a live legislative demand backed by public figures, adding pressure that could shape how fast other governments move on likeness rights.