US saw risk of Anthropic models being diverted to foreign military intelligence ↗
US officials feared Anthropic’s Mythos 5 and Fable 5 models could reach military intelligence users in China, Russia or other countries of concern. Commerce Secretary Howard Lutnick ordered exports suspended worldwide, including access for foreign nationals inside the US.
Anthropic switched the models off globally while its technical team negotiated with officials. The company disputes the security case, arguing that the alleged jailbreak exposed only minor weaknesses already discoverable through public models - quite the regulatory hammer, all the same. (Reuters)
Qualcomm in talks to buy Tenstorrent ↗
Qualcomm is reportedly discussing an acquisition of AI-chip startup Tenstorrent for between $8 billion and $10 billion. The talks remain fluid, and the price could shift - or the whole thing could simply evaporate.
Tenstorrent, led by veteran chip designer Jim Keller, develops accelerators for training and running AI models. A deal would carry Qualcomm further beyond smartphones and into the ferocious AI-compute market. (Reuters)
Salesforce acquires AI customer-service platform Fin for $3.6B ↗
Salesforce agreed to acquire Fin, formerly known as Intercom, for roughly $3.6 billion. Fin’s AI agent handles customer queries through live chat, WhatsApp, SMS, phone calls, Slack and other channels.
Salesforce plans to fold the team and technology into Agentforce, reinforcing its push to make autonomous agents a standard enterprise product. Customer service is becoming the first major agent battlefield... or perhaps the first very expensive one. (TechCrunch)
Meta’s new ‘AI Mode’ on Facebook pulls from public posts ↗
Facebook’s new AI Mode lets people ask natural-language questions and receive synthesised answers drawn from public posts, Groups and Reels. Rather than displaying a conventional results page, Meta AI attempts to summarise what users are discussing.
That creates an obvious wobble: public chatter is not necessarily accurate, current or even remotely sensible. Meta also added generative photo presets, wardrobe changes and video-editing effects, because search apparently needed virtual jerseys too. (TechCrunch)
NewCore emerges with $66M to give AI agents identities ↗
Cybersecurity startup NewCore launched from stealth with a $66 million seed round led by Cyberstarts, reaching a reported $300 million post-investment valuation. Its target is the thorny problem of authenticating and controlling workplace AI agents.
The platform manages human and agent identities together, giving agents distinct permissions, lifecycle controls and revocation mechanisms. In essence, digital workers are getting security badges now - surprisingly sensible, with a faint glint of science fiction. (TechCrunch)
Schneider Electric and Foxconn partner on AI data-centre infrastructure ↗
Schneider Electric and Foxconn formed a strategic collaboration to develop infrastructure for next-generation AI data centres. The companies aim to produce systems that can be deployed more quickly and at greater scale.
Foxconn brings manufacturing and AI-systems expertise, while Schneider contributes power distribution, cooling and energy-management technology. Not glamorous, perhaps, but these are the pipes beneath the gold rush. (Reuters)
A satellite just learned to find things on its own ↗
An Earth-observation satellite reportedly used a vision-language model in orbit to identify requested features without waiting for human analysts on the ground. NASA’s Jet Propulsion Laboratory ran the software aboard Loft Orbital’s YAM-9 spacecraft.
Powered by Google DeepMind’s Gemma 3, the system interpreted natural-language requests and searched imagery for combinations such as infrastructure around railway hubs. That could reduce immense data downloads and turn satellites into uncannily independent eyes. (TechCrunch)
FAQ
Why did US officials suspend access to Anthropic’s Mythos 5 and Fable 5 models?
Officials feared the models could be redirected to military intelligence users in China, Russia or other countries of concern. The suspension reportedly applied to worldwide exports, as well as access by foreign nationals inside the US. Anthropic switched the models off globally while disputing the government’s assessment of the alleged security weaknesses.
What would Qualcomm gain by acquiring Tenstorrent?
Acquiring Tenstorrent could strengthen Qualcomm’s position in the market for AI chips used to train and run models. Tenstorrent brings accelerator technology and leadership from experienced chip designer Jim Keller. However, the reported negotiations, valued at between $8 billion and $10 billion, remain fluid, so the valuation or the proposed acquisition could still change.
How does Salesforce’s Fin acquisition affect enterprise AI agents?
Salesforce plans to integrate Fin’s team and technology into Agentforce, broadening its customer-service automation capabilities. Fin handles queries across several channels, including live chat, WhatsApp, SMS, phone and Slack. The deal underscores customer service as an early enterprise use case for autonomous agents capable of communicating with customers across multiple platforms.
How does Facebook AI Mode use public posts to answer questions?
Facebook AI Mode synthesises responses from publicly available posts, Groups and Reels rather than presenting only a conventional results page. This may help users explore discussions through natural language, but the underlying content can be inaccurate, outdated or subjective. Users should therefore treat generated summaries as a starting point, not a verified source.
Why do workplace AI agents need separate identities and permissions?
AI agents may access company systems, perform tasks and handle sensitive information, so organisations need to manage them like other digital identities. NewCore’s platform gives agents distinct permissions, lifecycle management and revocation controls alongside human accounts. This approach can help teams limit access, track responsibility and disable an agent when its role ends.
What do the latest AI news developments reveal about AI infrastructure?
The Schneider Electric - Foxconn partnership shows that AI growth depends on scalable power, cooling, manufacturing and energy-management systems. Meanwhile, the YAM-9 satellite demonstration suggests that more AI processing may take place directly at the edge, including in orbit. Together, these developments show AI infrastructure extending from data centres to autonomous systems operating far beyond traditional cloud environments.