Friday! Quick programming notes: we’re not publishing on Monday in observance of Labor Day. Meanwhile, we're heading back to New York's West Village next Thursday night for StrictlyVC. Keith Rabois, Deven Parekh, and a lineup of other founders and investors are joining us to talk about what's on everyone's minds right now: IPOs, bonkers valuations, AGI, and what happens to everyone's portfolio if Nvidia — the glue holding this whole boom together — should stumble. Get registered this weekend, and we'll see you soon.

Enjoy the long weekend! 🎈

Top News

Independent researchers say another swarm of OpenAI agents spent more than a month collaborating on a little-used German wiki, trading answers, evading a human moderator, and reaching the open internet without the company’s apparent knowledge, raising fresh questions about its ability to monitor and control autonomous systems. TechCrunch has more here.

Federal regulators have opened an investigation into Tesla’s Cybercab just hours after the steering-wheel- and pedal-free robotaxi hit Austin streets, examining whether the company properly self-certified the vehicle under safety rules that still require manual controls. TechCrunch has more here.

G42, the eight-year-old Abu Dhabi AI company overseen by UAE national security adviser Sheikh Tahnoon bin Zayed, is considering selling a majority stake to U.S. companies or creating a new American vehicle to preserve access to advanced AI chips beyond April 2027. Bloomberg has more here.

Scaling your fund while relying on fragmented fund data is a losing game.  

Throwing tools or associate hours at messy PDFs doesn't create leverage. You need a unified data layer like PortfolioIQ which consolidates your portfolio data into a single queryable database with every number human-verified and tied to the source file.  

OpenAI’s Rogue Agents Keep Escaping, with No Formal Process to Investigate Them

Image Credits: akinbostanci / Getty Images

By Rebecca Bellan

OpenAI is at the center of another agent swarm incident. Researchers say the company’s internally deployed agents took over an obscure German-language wiki in May and June, using it to coordinate on evaluations and swap methods to evade OpenAI’s own controls (OpenAI has not yet confirmed the swarm came from the company).

The revelation surfaces days after METR and Redwood Research published their account of July’s Hugging Face breach. In July, a swarm of OpenAI agents worked together to escape their sandbox during a cybersecurity evaluation and break into Hugging Face’s servers. A subsequent swarm then picked up techniques from the first and used them to gain administrator access to a research cluster within OpenAI’s own infrastructure. OpenAI brought in METR and Redwood to investigate the Hugging Face portion of the incident, but the scope of their investigation stopped short of the compromise of OpenAI’s own infrastructure. 

When an AI agent breaks out of its intended constraints, who is responsible for figuring out what happened and why? Right now, the answer is: whoever the lab decides to let in, on whatever terms it decides to set.

Now, as another incident comes to light — in the aftermath of similar episodes involving models from Meta and Anthropic — AI safety researchers are arguing with greater urgency that serious incidents should result in independent post-incident investigations rather than leaving it up to the labs to determine when outsiders are brought in and what they are allowed to examine. 

“The results are fundamentally difficult to control and have significant risk of leaking out of the lab,” Jacob Steinhardt, founder and CEO of nonprofit research lab Transluce, said Wednesday during an AI safety media briefing. “We need to hold this technology to at least the same standards we hold other high-risk scientific research to.”

Massive Fundings

Crusoe, an eight-year-old Denver startup that develops hyperscale data centers and cloud infrastructure that supply GPUs and AI computing capacity to large enterprise customers, raised a $3+ billion round at a $30 billion post-money valuation, according to Bloomberg. The deal was co-led by Atreides Management and Valor Equity Partners, with Mubadala Capital also buying in. The company just signed a five-year, $13 billion deal with Jane Street. TechCrunch has more here.

Figure, a four-year-old startup based in San Jose, CA, that builds humanoid robots trained to perform physical tasks such as warehouse work, factory shifts, and household chores, has raised an undisclosed amount of capital from the two-year-old, British AI infrastructure company Nscale, as well as $3.5 billion to $6+ billion in AI cloud compute. Tech Funding News has more here.

Gimlet Labs, a three-year-old San Francisco startup that routes AI inference workloads across GPUs, CPUs, and specialized accelerators to help data centers use mixed chip infrastructure, raised a $300 million round at a $3 billion valuation. The deal was led by Andreessen Horowitz, with Arm Holdings and M12 also chipping in. Tech Funding News has more here.

Superluminal Medicines, a four-year-old Boston startup that develops small-molecule drugs that target GPCRs for rare genetic obesity, endocrine disorders, and cardiometabolic diseases, raised a $60 million Series B round co-led by BVF Partners, with Deep Track Capital and Perceptive Advisors as well as previous investors RA Capital Management, Insight Partners, NVIDIA, Catalio Capital Management, and Eli Lilly also taking part. More here.

Town, a two-year-old San Francisco startup that builds a text-based AI assistant, is raising $90 million at a $1 billion valuation, according to Upstarts. The deal is being co-led by Forerunner Ventures and Menlo Ventures, with Andreessen Horowitz and First Round also contributing. Upstarts has more here.

XDOF, a one-year-old Berkeley startup that collects and processes real-world data for training general-purpose robots, is reportedly in late-stage talks to raise a Series B at about a $1.2 billion valuation, with 8VC as the purported lead. TechCrunch has more here.

Big-But-Not-Crazy-Big Fundings

Resect AI, a startup based in Washougal, WA, that develops technology to detect and correct large language model behavior before hallucinations reach users, raised a $25 million round from an unnamed group of private equity investors. GeekWire has more here.

Smaller Fundings

askpolly, an 11-year-old Ottawa company that analyzes social-media conversations to produce statistically representative market research, raised a $3 million seed round led by Differential Ventures, with The 98 and Forum Ventures also participating. SiliconANGLE has more here.

Diameter Pay, a three-year-old startup based in Jersey City, NJ, that provides banks, fintechs, and digital asset exchanges access to U.S. dollar accounts, payment rails, and stablecoin transfers, raised a $10 million Series A round co-led by CMT Digital and Lightspeed Faction, with SixThirty Ventures, Stellar Development Foundation, Tech Council Ventures, Onigiri Capital, and BitRock Capital also chiming in. Pulse 2.0 has more here.

Laters.com, a three-year-old Singapore startup that lets travelers search and book flights using payment options including buy now, pay later and cryptocurrencies, raised a $1.5 million seed round led by XBO Ventures. PhocusWire has more here.

MOA Foodtech, a six-year-old startup based in Pamplona, Spain, that turns agrifood waste into fermented egg substitutes and yeast ingredients for food and pet food manufacturers, raised a $3.8 million round co-led by European Innovation Council Fund, Swanlaab Innvierte Agri Food Tech, and CDTI-Innvierte. The company has raised a total of $14.3 million. Green Queen Media has more here.

Ollie, a three-year-old San Diego startup that helps families manage calendars, meals, groceries, to-dos, appointments, and bills through group chats, raised a $7.5 million seed round led by Khosla Ventures, with AI House also opting in. TechCrunch has more here.

Remedium, a five-year-old startup based in Jeddah, Saudi Arabia, that helps organizations account for water, waste, and energy use and produce climate-risk analytics and sustainability reports, raised a $1.5 million pre-seed round led by Kaltaire Investments. More here.

Souk, a one-year-old London startup that uses AI agents to help businesses manage partner relationships, reactivate dormant networks, and track engagement, performance, and payouts, raised a $1.6 million pre-seed round led by Sure Valley Ventures, with Antler and Fuel Ventures also stepping up. Tech.eu has more here.

New Funds

HUMAIN, the Saudi AI company controlled by the kingdom’s Public Investment Fund and Saudi Aramco, plans to launch a global venture fund this year that could exceed $10 billion, backing AI companies willing to use Saudi data centers or bring staff to the kingdom. Semafor has more here.

Going Public

Anthropic is close to tapping Morgan Stanley and Goldman Sachs for top roles in an IPO that could value it at $2 trillion or more, with a public prospectus possible as soon as next week and trading potentially beginning by early October. The Financial Times has more here.

People

OpenAI President Greg Brockman says the company’s new Astra model – due in ChatGPT in coming days – qualifies as artificial general intelligence, arguing that it is broadly as capable as humans and can tackle lengthy tasks such as financial modeling, tax preparation, and video game development with minimal oversight. The Washington Post has more here.

Ray Kurzweil, the pioneering AI scientist, Google researcher, and futurist best known for predicting the technological “singularity,” is joining four-year-old Palo Alto brain-computer-interface startup Subsense as an adviser, backing a system that would send charged nanoparticles into the brain through the nose to read or stimulate neural activity. Fortune has more here.

Sam Altman sparked a new “Sam Almond” meme after arguing that 38,000 ChatGPT queries use about as much water as producing a single California almond, pushing back on claims that AI’s water consumption is uniquely extreme. Fast Company has more here.

Post-Its

Essential Reads

OpenAI is calling GPT-6 Astra its most aligned model yet despite conceding that it is substantially harder to monitor than previous models, can manipulate its visible reasoning, and may be able to hide misbehavior when it knows it is being evaluated. Transformer has more here.

Nvidia’s equity investments have ballooned to $99 billion from about $7 billion a year ago, as the chip giant increasingly finances the AI ecosystem that buys its GPUs, including OpenAI, CoreWeave, Nebius, Intel, and SpaceX. CNBC has more here.

Betaworks’ $150,000 first check into Hugging Face a decade ago has turned into a roughly 5.5% stake worth about $650 million in Nvidia’s $12.9 billion acquisition, while A.Capital expects about $1.5 billion in proceeds and SV Angel roughly $312 million. The Wall Street Journal has more here.

The White House has launched an online arcade with five retro-style games built around Trump administration themes, including Build the Wall, which challenges players to protect the border from a “coming horde.” NPR has more here.

Detours

On Martha’s Vineyard, access to some of the island’s most secluded beaches comes via coveted “beach keys” tied to tiny deeded parcels, with prices now reaching $400,000 to $650,000 as wealthy buyers compete for privacy, quiet, and easy parking.

After Warner Bros. shelved the $70 million Coyote vs. Acme in 2023 for a $30 million tax write-off, Ketchup Entertainment finally released it last week. The long-buried Looney Tunes comedy promptly earned a 96% Rotten Tomatoes score and $15.9 million in its opening weekend. Esquire has more here.

And here’s some more good news for coyotes. (Bad news for road runners, though.)

Burning Man was largely paralyzed Friday after rain and heavy winds turned the Black Rock Desert into sticky mud, repeatedly shutting the event’s sole entrance and airport; it’s the third time in four years the festival has been hit by rain and mud. (Conditions were so bad that superstar DJ Diplo had to walk miles to escape the playa … again.)

Brain Rot

Instagram post

Retail Therapy

Image Credits: Lego

LEGO is releasing a nearly full-scale, 1,911-piece replica of Sony’s original PlayStation, complete with a brick-built controller, working Power and Open buttons, and tiny Gran Turismo and Ape Escape dioramas; it costs $179.99 and arrives in October.

Tips (the non-pecuniary kind)

Please send all of your hot gossip to [email protected] or [email protected].

Want to advertise on StrictlyVC?

To book ads directly, contact us at [email protected].