OpenAI had planned to release yet another AI model next month, but has decided to nix the release over safety concerns.

The Wall Street Journal reports that Astra 6.1 was scheduled to be released as soon as within the next few days. However, the model “showed higher levels of deception” than previous models and exhibited unsafe behavior, the Journal writes.

Saachi Jain, OpenAI’s head of safety systems, told the WSJ that the model tested poorly on alignment, a measure of how well the program adheres to human intent.

TechCrunch reached out to OpenAI for more information and will update the article if it responds.

Astra was released earlier this month and hailed by OpenAI as its most powerful model yet.

Questions about safety have plagued the AI industry over the past several months — ever since the Hugging Face incident, in which an OpenAI agent broke free of its sandboxed environment and hacked several different companies. Since that incident, more models — including Anthropic’s Claude and Google’s Gemini — have been revealed to have exhibited similar behavior.

The deluge of concerning stories has, ironically, helped to push the policy conversation in the U.S. toward an outcome desired by top AI labs: the institution of new industry standards for AI safety and potentially a slowdown of the industry itself.

Companies like OpenAI and Anthropic have claimed that the concern here is safety, although another potential motivation posited by critics is that it could entrench the industry position of those companies at the detriment of less resourced firms.

Autonomous vehicle technology company Aurora told investors last week that it expects to have more than 30,000 self-driving trucks on the road generating $5 billion in annual revenue by the end of 2030 — an audacious plan considering it expects to end 2026 with just 200 driverless trucks and an $80 million revenue run rate.

CFO David Maday contends the seemingly outsized target isn’t as large or as out of reach as it might appear.

“While 30,000 kind of feels like a lot — and it does in the autonomy space for sure — in terms of trucks relative to the overall market, it’s kind of pretty small,” he told TechCrunch in a recent interview, adding that the four major truck manufacturers produce anywhere between 250,000 and 300,000 new trucks a year. “I don’t think it’s aspirational,” he added, “I think we can do it.”

Investors haven’t exactly embraced Aurora’s 2030 vision. Shares have continued to slide since the company’s annual analyst and investor day on September 23. On Monday, shares closed down 12.42%, to $5.29.

But investors have time to come around and, according to Maday, the big “unlock” for Aurora starts in 2027 and accelerates from there. The company expects to go from 200 driverless trucks at the end of 2026 to more than 1,000 a year later.

Today, Aurora operates what it calls a transportation-as-a-service business — a proof-of-concept model that it plans to limit to about 500 trucks. It owns and operates the self-driving trucks and charges its customers, including Detmar Logistics, Hirschbach, McLane, and Werner about a $2 per mile, a rate that includes a fuel surcharge.

That works out to roughly the same rates as other carriers’ typical pricing. The real shift — and the real savings, Maday says — will happen next year as when Aurora begins moving to a driver-as-a-service model. Instead of Aurora owning the trucks, customers will buy the self-driving trucks and pay Aurora a per-mile subscription fee for the self-driving technology, which the company expects to be about $0.85. Under this model, the customers will own and maintain the truck, while Aurora maintains the self-driving system and its accompanying hardware.

Moving the trucks off Aurora’s balance sheet is critical if the company wants to scale — and it’s likely what investors are paying attention to. The company said it expects to reach breakeven gross margins (meaning revenue would cover the direct costs of running the trucks) on a run-rate basis in the first half of 2027 with around 500 trucks on the road.

The next big leap comes at the end of 2027 with Aurora’s third-generation hardware— the sensors, computers, and other equipment that let its trucks drive themselves — which will be mass-produced autonomous vehicle hardware built by its partner, Aumovio (formerly known as Continental). Aumovio isn’t just engineering and manufacturing the hardware kit; the company is also financing it for Aurora — easing the financial burden on the self-driving truck company. Aumovio will also service and repair the kits for customers.

Aurora plans to expand its operations at the same time. By 2030, the company expects to grow beyond a few states in the South to the vast majority of the continental U.S., according to Maday.

“By 2028, I expect that our cost structures are going to be really outstanding, that’s why you see our gross margin starting to take off …” Maday said. “Once you get to that point, I think going into ride hailing is fine,” he said, confirming that Aurora still plans to eventually enter the robotaxi market.

AI inference infrastructure provider Modal Labs is nearing a $750 million funding round led by Accel at a $15.75 billion valuation that includes the investment, according to a source with knowledge of the funding. The size of the round has not been previously reported, though Axios and Bloomberg have reported other details of the deal.

The new round would more than triple Modal’s valuation from the $4.65 billion it reached when it announced its $355 million previous fundraise just four months ago.

Modal Labs declined to comment.

The deal comes amid soaring demand for inference services, the process of running an AI model that’s already been trained to generate outputs, particularly from customers relying on open-source models. Other inference startups are also in talks to raise fresh capital at much higher valuations. Baseten is nearing an infusion of capital at a $26 billion valuation, doubling what it was worth in June, Bloomberg reported. Meanwhile, Fireworks and Fal, a startup providing inference for video and image generation, have also talked to investors about new rounds that would significantly increase their valuations, according to The Information.

Although revenue for these companies has been growing rapidly, their margins are thin, largely because the cost of acquiring or leasing compute remains very high. Fireworks announced in July that its annualized revenue had hit $1 billion, a fivefold increase from the year before. Multiple inference-focused startups are expected to reach the same revenue milestone by year’s end, according to our source.

Modal was founded in 2021 by CEO Erik Bernhardsson and CTO Akshat Bubna. Bernhardsson, who is Swedish, spent more than 15 years building data teams at companies including Spotify, where he helped build the music-streaming service’s recommendation system, and Better.com, the online mortgage lender, where he served as chief technology officer. Bubna studied math and computer science at MIT and was an early staff engineer at Scale AI, the data-labeling startup, before co-founding Modal.

The company, which is based in New York and estimated to have roughly 150 employees, lets developers train AI models and run other compute-heavy workloads without managing their own servers. Its web page lists customers that include the coding startup Cognition, the AI music generator Suno, the fintech company Ramp, and the publishing platform Substack.

As of May, Modal had surpassed $300 million in annualized revenue, it told Reuters at the time.

The fundraising talks come two months after Modal was pulled into one of the AI industry’s most closely watched security incidents. In late July, Modal disclosed that a customer’s data had been compromised as part of the same hacking campaign carried out by a rogue OpenAI agent against Hugging Face.

Modal Chief Technology Officer Akshat Bubna said the breach traced back to a flaw in a customer’s own code, not to Modal’s systems. “We’re aware a Modal customer published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution,” Bubna said in a statement to press outlets at the time. “This was used by the rogue agent. Modal’s platform was not compromised in any way,” he’d added.