OpenAI●Covered by 9 sources
OpenAI says it scanned 30M+ commits and found 500K+ fixes, and now wants to land them too
OpenAI expanded its Daybreak cybersecurity program with a Codex Security plugin, the full GPT-5.5-Cyber model (which it claims beats Anthropic's Mythos on a benchmark, by OpenAI's own measure), and a Patch the Planet initiative aimed at open-source maintainers, with a partner network of 25+ security firms and several governments. The pitch has shifted from finding bugs across 30K+ codebases like Python, Go, and cURL to automatically writing and shipping the patches.
So what? ↓Most of the software running your bank, your phone, and your hospital sits on volunteer-maintained open-source code, so if an AI can actually close those holes faster than attackers exploit them, that's the unglamorous fix that protects you, but "outperforms" on the vendor's own scorecard still needs an outside referee.you your your your.running your your your your your your your your your your your your your.running.your your your your.running.your your your.running.your your your your.running.running.
Source: OpenAI
Valve●Covered by 6 sources
Valve's Steam Machine lands at $1,049, and you'll need to win a lottery to buy one
The 6-inch cube ships June 29 with 16GB of DDR5 RAM and 512GB of storage (no controller in the box), scaling to $1,349 for the 2TB version, and Valve says it's selling basically at cost. To buy one you get dropped into a randomized purchase queue, which Valve frames as making things "less frustrating and more fair."
So what? ↓The price tag is nearly double a PS5 because this is a tiny PC, not a console, so the real question isn't whether it's expensive, it's whether you'd rather just install SteamOS on hardware you already own and skip the lottery entirely.
Source: Valve
Google DeepMind●Covered by 6 sources
Google puts $75M into A24 to build AI filmmaking tools
Google DeepMind is investing roughly $75M in indie studio A24 for a multiyear partnership building AI movie-production tools, including storyboard workflows pitched as filmmaker-shaped rather than full prompt-to-movie generation, and notably without Google getting access to A24's data. The studio behind Hereditary and Everything Everywhere All at Once is now also the studio testing whether AI belongs on set.
So what? ↓A prestige studio with arthouse credibility signing on gives AI filmmaking a respectability stamp that cheaper, sloppier tools could never buy, which is exactly how the technology creeps from novelty into the actual movies you pay to see.
Source: Google DeepMind
Sakana●Covered by 5 sources
Sakana's Fugu doesn't build a smarter model. It hires a manager for the ones you already have
Japan's Sakana AI launched Fugu, an OpenAI-compatible API that routes each request to whichever specialized model should plan, execute, verify, or answer it, with Sakana claiming both versions land at or near Anthropic's Fable 5 and Mythos on coding and reasoning. Vercel already added Fugu Ultra to its AI Gateway, but reviewers are flagging opaque baselines, no cost accounting, and shaky comparisons, and early performance is reported as mixed.
So what? ↓The pitch underneath the orchestration is dodging dependence on any single provider (and the export controls that hit Anthropic's top models), which matters because the alternative is your whole AI stack going dark the day one company's chips or politics get cut off.
Source: Sakana
Hugging Face●Covered by 5 sources
GLM-5.2 is now the best open model, and it ranked #3 overall on GDPval-AA
GLM-5.2 landed as the strongest openly available model, the first open-weight one widely treated as good enough for agentic work, with distributions on AWS Marketplace, Baseten, Fireworks, and 20+ other providers. It's still text-only (no vision), so in a head-to-head building a 3D WebGL game, Claude Opus came out faster and cleaner while GLM-5.2 was cheaper but rougher with no way to visually check its own work.
So what? ↓An open model you can run locally that's now sitting at #3 overall means the gap between the free, own-it-yourself tier and the pricey closed labs keeps shrinking, which eventually shows up as leverage on whatever you're paying OpenAI or Anthropic.
Source: Hugging Face
SpaceX●Covered by 5 sources
SpaceX is quietly running a $28B GPU rental shop, and just signed a $6.3B tenant
SpaceX's Colossus data centers, originally built to train Grok, just leased up to $6.3B of Nvidia GB300 compute to open-model startup Reflection AI, the third such deal after Anthropic and Google. That pushes its compute-leasing run rate to roughly $28B annualized, about double CoreWeave's, even as a $20B-$25B debut bond sale reportedly drew $89B in demand.
So what? ↓The rockets get the headlines, but the business funding your satellite internet and the next AI boom is increasingly a landlord renting out graphics cards by the hour.
Source: SpaceX
NCSC (UK National Cyber Security Centre)●Covered by 4 sources
Five Eyes says cyber-attack-capable AI is "months" away, not years
The intelligence agencies of the US, UK, Canada, Australia, and New Zealand jointly warned that frontier AI is compressing cyber risk timelines from years into months, with models capable of major attacks on governments and businesses reportedly close at hand. The agencies are telling executives to harden defenses now, as White House talks pivot toward shared AI security benchmarks.
So what? ↓When five spy agencies agree the clock is measured in months and tell your company's bosses to act, the breach that empties your account is no longer a someday problem but a this-quarter one.
Source: NCSC (UK National Cyber Security Centre)
TechCrunch●Covered by 4 sources
A hack at a company you've never heard of just leaked data from LastPass, Jamf, and HackerOne
Market intelligence firm Klue was breached via a compromised legacy credential tied to a cloud integration tool, and the cybercrime crew Icarus is taking credit for hitting what The Register says is hundreds of victims. LastPass has confirmed customer personal info and support case records were stolen, its second breach in recent years to come through a partner rather than its own systems.
So what? ↓You can pick the most paranoid password manager on earth and still get exposed because some vendor you never signed up with left an old login lying around.
Source: TechCrunch
Chevron●Covered by 3 sources
Microsoft just locked in 20 years of gas-powered emissions to run a Texas data center
Microsoft signed a 20-year power purchase agreement with Chevron for a new natural gas plant feeding a West Texas data center, one of the largest gas-powered data center projects in the US. The same company that has spent years touting carbon-neutral pledges is now committing to decades of fossil-fuel emissions to keep the servers humming.
So what? ↓A 20-year gas contract is a long time to burn fossil fuel for a company that loves its climate pledges, and it's a concrete reminder that the power running these data centers has to come from somewhere real.
Source: Chevron
techmeme.com●Covered by 3 sources
Anthropic just put an always-on Claude inside your Slack, one channel at a time
Claude Tag is a new agentic AI "coworker" that lives in Slack, learns your channels' context, and offers suggestions, currently in beta for Claude Team and Enterprise tiers, with each channel able to run its own isolated Claude identity. TechCrunch reads the real move as Anthropic vacuuming up your company's institutional knowledge and workflows, not just shaving minutes off a thread.
So what? ↓An AI that quietly learns every channel it sits in is also an AI that learns exactly which desk jobs a company could stop paying for, starting, per Engadget, with Slackbot's.
Source: techmeme.com
MIT Technology Review●Covered by 3 sources
Anthropic spent years warning AI is dangerous, now the government may take the hint and ban exports
Anthropic's clash with the US government traces back to April, when it announced an AI model called Mythos and dispatched a hacker to reassure officials about safety. Its long record of sounding louder alarms about advanced AI than rival OpenAI may now be feeding momentum toward export restrictions.
So what? ↓When the company that builds the thing keeps insisting the thing is scary, regulators eventually believe it, and the rules that follow shape which AI tools you can actually use.
Source: MIT Technology Review
Netflix●Covered by 2 sources
Netflix's new horror game makes your phone the controller and Zoë Kravitz the hands
Netflix launches Unhinged on June 30th, a cloud-based TV horror game from Night School Studio (of Oxenfree fame) that uses your smartphone to control star Zoë Kravitz's on-screen hands. It's a sharp turn from the trivia, party games, and oddly outdated FIFA that have defined Netflix's TV gaming push so far.
So what? ↓This is Netflix testing whether the second screen in your hand can turn its passive lean-back gaming into something you actually steer, the difference between a gimmick and a reason to keep the subscription.
Source: Netflix
theneuron.ai●1 source · panel-picked
GPT-5.6 reportedly lands June 25 with a 2M-token context window
OpenAI is rumored to ship GPT-5.6 on June 25 with a 2M-token context window, cheaper pricing, better agentic coding, image-to-code replication, and Playwright-style browser testing baked into ChatGPT. Every word of that comes from the rumor mill, so hold it loosely until OpenAI actually confirms anything.
So what? ↓The cheaper pricing and stronger coding chops are the parts worth wanting, but "reportedly" is carrying this entire paragraph, so file it under wish list until there's a launch you can actually click.
Source: theneuron.ai