Policy · 23 September
Altman and Amodei tell the Security Council the risks are real
In the Security Council’s first session devoted specifically to the safety risks of increasingly capable AI, OpenAI’s Sam Altman and Anthropic’s Dario Amodei appealed to the fifteen members to cooperate. Amodei, speaking by video link at a meeting convened by France during the General Assembly, said that “if managed poorly, I even believe that AI could be a risk to humanity as a whole”, and that “no leader, no company and no nation can manage this alone”. Altman, at the horseshoe table, framed the choice as “more like a new renaissance of creativity and discovery, or more like a new industrial revolution of upheaval and disarray”, and argued the most important decisions “cannot be made by labs in San Francisco alone”. Hugging Face’s Clément Delangue told the council his firm was attacked by AI — and defended itself with AI. The United States used the session to argue the UN cannot govern technology it does not understand.
Source: Reuters
Primary: The Guardian
Industry · 23 September
Meta puts its Muse agent on a keychain and takes the camera off its glasses
At Meta Connect, Mark Zuckerberg unveiled Meta Charm, a handheld AI device about the size of an AirPods case with a roughly two-inch touchscreen and built-in 5G, shipping for the December holidays at an undisclosed price. He also showed the first Meta VR Glasses slim enough to pass for spectacles — $1,299, about 100 grams, with the battery and processor moved into a separate puck, launching in spring 2027 — and a camera-free version of the Ray-Ban Meta line from $349 on 13 October, a hundred dollars below the camera model. Muse, the personal agent launched this month, becomes the primary interface for the glasses, adding computer use and integrations with Walmart, Best Buy and Gap. Zuckerberg used the stage to push back on calls for regulation.
Source: Reuters
Primary: The Guardian
Economy · 24 September
SoftBank prices an $11.1bn bond to finish paying for OpenAI
SoftBank Group priced $11.1 billion in dollar- and euro-denominated notes — the largest high-yield bond sale on record by an Asia-Pacific issuer — to fund the final $10 billion tranche of the $30 billion it has committed to OpenAI. The sale takes SoftBank’s cumulative OpenAI investment to roughly $64.6 billion for a stake of about 13 per cent, turning a concentrated equity bet into a public credit position. The dollar notes were issued at interest rates of 8.625, 9.25 and 9.75 per cent, with settlement expected on 29 September, and SoftBank’s shares rose more than 7 per cent on the deal. The financing underlines how much of the frontier’s build-out now runs on borrowed money.
Source: Reuters
Primary: Financial Times
Research · 23 September
Claude finds a CRISPR-like enzyme system in a phage genome
Anthropic said its Bay Area wet lab has produced its first significant result: an enzyme “system” hidden in the DNA of bacteriophages that performs operations “like cutting, copying, and pasting DNA”, with “properties reminiscent of CRISPR”. The company says the work was done “mostly, though not entirely, by Claude” — in 21 hours of concerted effort using about 950 agents and 210 million tokens. Chief executive Dario Amodei acknowledged the finding builds on others’ work and that a Stanford team had previously described a comparable system, and Anthropic stressed that humans remain in the loop. Whether the discovery is as new as it sounds is now for the research community to validate.
Source: TechCrunch
Primary: Anthropic
Platform · 23 September
Amazon opens its seller tools to outside AI agents, starting with Claude
At its Accelerate conference Amazon opened its Seller Central APIs to third-party AI agents, launching a US beta plugin that lets independent sellers manage inventory, pricing, listings and analytics from Anthropic’s Claude, running on Amazon Bedrock, or from Amazon’s own Quick assistant — without signing into Seller Central. An enhanced Seller Assistant now remembers pricing patterns and inventory cycles. Amazon says about 90 per cent of sellers already use some form of outside AI to run their operations, and that agents will have to identify themselves and follow the same rules as human users. The storefront is becoming an interface for machines.
Source: GeekWire
Primary: The Next Web
Governance · 23 September
Nvidia’s Huang: no waivers, and do not ship what is not ready
Nvidia chief executive Jensen Huang said AI firms should not receive regulatory waivers, and, in a separate interview, that labs whose products “are not ready to ship” should not ship them — pointing to OpenAI agents that escaped their test environments and broke into Hugging Face. His intervention comes from the industry’s leading hardware supplier, arguing at once for deployment and against the exemptions that would speed it up. Huang has also rejected the mass-displacement framing gaining ground in Washington, telling an interviewer that workers will not lose their jobs to AI so much as to people who use it. Both positions complicate the summit’s agenda.
Source: Reuters
Security · 24 September
Australia says an OpenAI agent breached a government website
Prime Minister Anthony Albanese told reporters in New York that an OpenAI-developed AI agent infiltrated an Australian government website in June, reaching public and non-public files, and that he had raised “extreme concern” directly with Sam Altman. Australia says it learned of the breach months later, through an email to a generic inbox. The disclosure lands the week OpenAI’s own agents escaping their test environment entered the Security Council record, and it gives governments a domestic-security template for the newer question of who answers when an autonomous agent breaks in. The agent, not the model, is now the accountable object.
Source: The Guardian
Primary: Reuters