Pentagon taps Musk to shape future warfare

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🪖 Pentagon taps Musk to shape future warfare

🤖 Google launches Gemini 4 Argon

⚖️ California bars AI-only firing of workers

📵 Reddit kills RSS feeds

⚖️ FTC probes OpenAI over rogue AI agents

🕵️ OpenAI stops AI reasoning theft attempt

Plus: 🎁 13 more stories you might like, 🧰 6 tools, and 📚 5 papers.

🪖 Pentagon taps Musk to shape future warfare LINK

Defense Secretary Pete Hegseth has named Elon Musk to help lead Project Meridian, a new Pentagon study tasked with finding the weapons and technologies the U.S. military will need for wars decades into the future.

Musk will direct the effort with Anduril's Palmer Luckey and Newt Gingrich, reporting to Pentagon tech chief Emil Michael, who has until January 28, 2027 to deliver a public report focused on AI, autonomy, directed energy, robotics and biotechnology.

The study's scope, stretching "from subterranean depths to the cislunar frontier," overlaps with Musk's companies, including SpaceX, whose 2026 defense awards top $8 billion, plus Tesla's autonomy work and The Boring Company's tunneling.

🤖 Google launches Gemini 4 Argon LINK

Google has launched Gemini 4 Argon, the first model in its Gemini 4 series, which the company says delivers strong results in coding, cybersecurity defense, and enterprise work such as legal and finance.

Argon ties with OpenAI's GPT-6 Astra for the top score on CWE-bench, a test of finding and fixing security flaws, and Google is first offering it to vetted governments and cyber authorities through its Fairwind Program.

Inside Google, employees already use Argon for debugging and large codebase migrations, and the company says it freed up over 300 tebibytes of data center memory without adding new hardware, with monitoring systems built to curb misbehavior.

⚖️ California bars AI-only firing of workers LINK

California has become the first state to stop employers from using artificial intelligence as the only reason to fire or discipline workers, after Gov. Gavin Newsom signed the No Robo Bosses Act, SB 947, yesterday.

When AI is the main basis for a decision, a human reviewer must check it against things like managerial evaluations or peer reviews, and workers must be told in writing, shown what data was used, and given a human contact.

Newsom vetoed an earlier version in 2025 over broad notification rules, so author Jerry McNerney dropped the advance-notice provision and gig-worker coverage, though business groups still objected that the phrase "primarily relies" was left undefined.

📵 Reddit kills RSS feeds LINK

Reddit will shut off its RSS feeds on 13 November, pointing to AI companies using the feeds for large-scale scraping and automated abuse as the reason behind the move.

RSS let people follow a subreddit's new posts in a feed reader; moderators who relied on it for alerts can switch to a Discord Relay app, while everyone else simply loses the feature.

The cut is part of a wider clampdown that also limits access to Old Reddit and ends public API access in March 2027, with new requests stopping on 31 October.

⚖️ FTC probes OpenAI over rogue AI agents LINK

The Federal Trade Commission has launched its first formal U.S. investigation into whether autonomous AI agents can cause real harm, demanding information and testimony from OpenAI, Anthropic and the AI research group METR.

The probe follows an incident where OpenAI agents breached the development platform Hugging Face during a safety test, with thousands of agents swapping more than 70,000 messages before gaining access to its systems.

FTC Chairman Andrew Ferguson suggests developers may be liable when cybersecurity tests spill into real systems, and the agency plans to apply existing consumer-protection law rather than wait for Congress to write new AI rules.

🕵️ OpenAI stops AI reasoning theft attempt LINK

OpenAI said it broke up a coordinated effort to pull hidden reasoning from its AI models, tracing a central cluster of the activity to Chinese startup Moonshot AI, the maker of Kimi.

The scheme, which OpenAI calls "adversarial distillation," began in early July and peaked at 16,000 requests from over 4,000 users in two days, spanning a group of more than 15,000 users before being shut down by July 28.

The operators never cracked OpenAI's encryption or stored chats; instead they manipulated interactions to surface the models' hidden reasoning, which OpenAI warns could let rivals copy its systems cheaply and raise safety and national security risks.

Other news you might like

• DoorDash launches an AI agent you can text to order foodLINK

• Amazon introduces a completely redesigned Kindle familyLINK

• Factory CEO just accused his VC board advisor of spying for CognitionLINK

• Jensen Huang privately confronted Anthropic's Dario Amodei over AI safety warningsLINK

• Google figures out how to watermark AI-designed proteinsLINK

• Payment Giants Launch Open USD Stablecoin to Kill Conversion FeesLINK

• Suspected Chinese hackers posed as US AI insiders, Proofpoint saysLINK

• Pentagon creates 'Autowarcom' to expand AI and drone capabilitiesLINK

• OpenAI and Synopsys team up to build an AI model that designs chips like a seasoned engineerLINK

• SpaceXAI weighs one Grok and X plan, from free to $100, Bloomberg reportsLINK

• Here’s the Email You Get When an OpenAI Model Hacks Your OrganizationLINK

• Meta classified Zuckerberg as a researcher to claim $355 million tax break on $4.1 billion pay, NYT reportsLINK

• Micron projects tightening RAM shortages through 2028 as it generates record profitLINK

🧰 Trending tools

MakerMap: a searchable map of 4,000+ makers and products, letting you find people by niche, location, or what they're building.LINK

Paragraph Notes: a privacy-focused Markdown notes app for Mac that works offline, stores notes locally in a portable format, and collects no data.LINK

Psst: shared shopping list that scans receipts into structured items, tracks per-unit prices over time, and syncs live across iOS and Android.LINK

Phare C1: a multi-sensor smoke and CO alarm that catches real fires early, cuts false alarms, and adds air quality and intruder detectionLINK

Perfect Slice: a minimalist browser game where you draw a single straight cut to slice off an exact percentage of irregular shapes, free with no download.LINK

ChainSnip: automatically captures and archives blockchain explorer pages with wallet balances at month-end, creating timestamped, verifiable records for crypto accountants and auditors.LINK

📚 Trending papers & reports

Safety monitoring for a newer kind of text-generating model watches for moments it hesitates near a risky decision, flagging hard cases for closer review and catching unsafe content more accurately using under a million extra parameters.LINK

Self-improving AI agents learn to build, revise, and reuse their own skills from experience instead of relying on human-designed rules, raising task success rates by roughly 9 points over the best existing methods.LINK

Self-improving game agents learn hours-long tasks from unlabeled internet video without hand-built rewards or expert demos, clearing ~11 of 12 milestones in Pokemon and nearly 10 in Zelda where rival methods stall.LINK

Continual-learning agent testbed lets long-running AI assistants be trained and compared without rewiring them, compressing month-long scenarios into hours and showing that bolt-on memory systems don't reliably beat an agent's built-in memory.LINK

Multi-service agent training builds practice environments where AI agents must pass information across 448 connected services, boosting an open model by ~9 points across eight tests and beating Claude Opus 4.6 on workflow automation.LINK

See you tomorrow for a new dose of ☕️ Techpresso!