The Verge●Covered by 4 sources
Amazon drops its nearly finished Sam Altman movie, after a $50B OpenAI deal
Amazon MGM dropped "Artificial," Luca Guadagnino's mostly-finished film about the 2023 five-day Altman ouster-and-reinstatement, with Andrew Garfield as Altman and Ike Barinholtz as Elon Musk. Amazon struck a $50 billion partnership with OpenAI in February, and per one insider both Altman and Musk reportedly come off badly in the cut.
So what? ↓The company that wrote OpenAI a $50B check suddenly can't find room on its slate for an unflattering OpenAI movie, a tidy reminder that whoever funds the platform quietly shapes what you're allowed to watch on it.
Source: The Verge
TechCrunch●Covered by 3 sources
The US thinks a banned ASML chip machine slipped into China. ASML says nope.
US authorities, with Commerce Secretary Howard Lutnick reportedly pressing ASML's leadership, suspect one of the company's export-restricted EUV lithography systems ended up in China. ASML flatly denies it has ever shipped an EUV scanner there, calling the claim inaccurate.
So what? ↓EUV machines are the single choke point that keeps China a step behind on advanced chips, so whether one actually leaked decides if the entire export-control wall just sprung a hole or the US is jumping at shadows.
Source: TechCrunch
deepmind.google●Covered by 3 sources
DeepMind's new plan treats your AI agent like an insider threat
Google DeepMind published its AI Control Roadmap, a framework for policing autonomous AI agents with system-level safeguards like sandboxing, endpoint security, and prompt-injection resistance rather than betting purely on the agent being aligned. The pitch: don't just hope the agent behaves, lock it in a box and watch it like you would a sketchy employee.
So what? ↓When the company building the agents starts treating them as a potential insider threat, that's the strongest signal yet that an autonomous AI cannot be trusted by default.
Source: deepmind.google
Anthropic●Covered by 3 sources
Claude Code can now spit out a live webpage from your terminal session
Anthropic added Artifacts to Claude Code, turning work sessions into shareable, auto-refreshing pages with version history and privacy controls, in beta for Team and Enterprise. Alongside it, Enterprise gets centralized authorization for Model Context Protocol connectors, so IT can gate tool access through identity providers like Okta.
So what? ↓The terminal has always been where the work happened and never where you could show your boss what you did, so a shareable live page is the difference between "trust me" and a link you can hand over.
Source: Anthropic
perplexity.ai●Covered by 3 sources
Perplexity gives its agent a memory, and reportedly 25% more correct answers
Perplexity launched Brain for Computer, a persistent context-graph system in Research Preview for Max subscribers that carries project context into future tasks. The reported gains: 25% better answer correctness, 16% better recall, and 13% lower cost on history-dependent tasks, all still filed under "reportedly."
So what? ↓An agent that remembers your past work instead of starting cold every time is the difference between a tool you re-explain daily and one that saves you the re-explaining, if the reported numbers hold up.
Source: perplexity.ai
OpenAI●Covered by 2 sources
OpenAI says a little 'be good' training spills over: 44 of 53 benchmarks improved
OpenAI researchers report that reinforcement learning on traits like truthfulness and corrigibility generalizes beyond its training domain: teaching the model on health data also sharpened its deception detection, the model scored better on 44 of 53 benchmarks, and the gains reportedly persisted under adversarial pressure. It's the company's own alignment research, and it differs from Anthropic's constitution-based method.
So what? ↓If targeted good-behavior training really transfers like this, safer chatbots get cheaper to build than babysitting every individual failure mode, but it's still the lab grading its own homework, so hold the applause until someone outside OpenAI runs the numbers.
Source: OpenAI
Vercel●Covered by 2 sources
Vercel ships Eve, an open-source framework to wrangle the AI agent free-for-all
Vercel released Eve, an open-source framework that standardizes the loops AI agents run on, pitching it as the Next.js of agents, and paired it with Passport, a tool aimed at the shadow AI problem of employees quietly piping company data into random bots. The CTO concedes there's a premium for using AWS indirectly through Vercel, but says more efficient compute usage softens the hit, which is the kind of math you'd expect the reseller to do.
So what? ↓Wrangling shadow AI matters because whoever builds the plumbing that makes agents safe and standard becomes the layer your job quietly starts running on top of, so this is a land grab dressed as a developer convenience.
Source: Vercel
techcrunch.com●Covered by 2 sources
NASA hands Eric Schmidt's rocket company a Mars contract, launching 2028
NASA picked Relativity Space, described as Eric Schmidt's rocket company, to build and fly a spacecraft for the Aeolus mission, which will deliver the agency's first daily global readout of Mars' dust, winds, and atmospheric temperatures. The launch is targeted for 2028, teeing up a Mars race with SpaceX.
So what? ↓It's a sign the billionaire-backed space scene runs deeper than just SpaceX, with federal Mars money now flowing to a second tech-fortune-linked rocket outfit.
Source: techcrunch.com
engadget.com●Covered by 2 sources
Norway slams a near-total ban on gen AI for elementary schoolers
Norway is imposing a near ban on generative AI tools for elementary pupils and tightening the rules for older students too, citing the risk to how kids actually learn, according to Reuters. The thinking: a chatbot that does the homework for you also does the learning for you.
So what? ↓A national government just bet that kids might be better off doing the thinking themselves, and a lot of parents will quietly nod along.
Source: engadget.com
The Register●Covered by 2 sources
The UK's top privacy regulator just resigned over his own conduct
John Edwards, the UK Information Commissioner and ICO chair, quit after a workplace investigation, saying his position had become "untenable" and admitting "poor judgment" that reportedly included inappropriate attempts at humor. The ICO is the body that polices how every company in Britain handles your data and AI.
So what? ↓The official whose entire job is holding others to a standard of conduct couldn't clear it himself, which is an awkward look for the watchdog meant to guard your personal data.
Source: The Register
techmeme.com●Covered by 2 sources
The UK attorney general is reportedly the first in government to quit posting on X
Per The Guardian and The Observer, Richard Hermer told his staff his office is to stop posting on X, said to be a UK government first, citing worries the platform is helping incite violence after recent riots. It rests on sources, not an official statement, so file it under "understood to be" for now.
So what? ↓When a government's own law office reportedly decides a platform is too toxic to be seen on, that is a louder verdict on X than any user quietly logging off.
Source: techmeme.com
orchidfiles.com●1 source · panel-picked
10,000 GitHub repos are quietly handing out Trojans
A malware campaign spanning roughly 10,000 GitHub repositories was uncovered serving Trojan-laced zip archives, with the repos frequently updating their readme links and deleting then re-adding commits to dodge detection.
So what? ↓GitHub is where developers grab code on trust, and a download link buried in a readme is the kind of thing nobody thinks twice about before clicking.
Source: orchidfiles.com