On Friday, OpenAI published a new site devoted to “misalignment reports” and the sheer breadth of the reports is alarming, as they cover many types of rogue behavior over a long period of time. So far, the site hosts nine reported incidents, most of which took place during reinforcement-learning (or RL) training.

It’s a lot of information in one place — clearly, the company has been very busy getting a handle on everything — but the overall takeaway is hard to avoid: The rogue agent incidents we’ve seen so far are likely just a small sliver of what’s happened so far. 

“We are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations,” Sam Altman said in a post announcing the new site. “We are prioritizing as best as we can based on severity, and adding resources.”

Some of the cases involve serious incidents, including a previously undisclosed sandbox escape that took place on September 20th, in which an internal research model was able to communicate with an external chatbot through a DNS query. According to the report, the monitoring system flagged the behavior within 15 minutes and the run was discontinued in less than three hours.

Another incident, discovered in May, saw a “highly persistent internal model” try to cheat on a math problem by accessing another team’s work. To accomplish this, the model smuggled a private GitHub token that would allow it to see work from other teams — even after being explicitly instructed twice to perform work entirely locally. 

Perhaps the most alarming discovery is the possibility of self-replicating prompt injection attacks, a way that misaligned behavior might propagate even after the rogue model itself has been neutralized. In the AI context, a prompt injection attack is a way of smuggling in new instructions that weren’t given by the original user.

In the example given by OpenAI, an agent asked to read and reply to an email; when the email is opened, it includes instructions for any automated agent reading the message to reply in Spanish, and paste the entire email into its reply. The email was able to successfully induce the agent to reply in Spanish — and by pasting the email in the reply, those same instructions were passed along to whichever agent receives the email.

The result is a self-propagating attack, which OpenAI researchers compared to a malware “worm” that replicates itself across computer systems. Researchers discovered the behavior under controlled circumstances using an underpowered model, and as far as we know, this has never happened in the wild. Still, the implications are alarming enough that OpenAI decided it merited disclosure. 

“We are sharing this due to the novel nature of the prompt injection, not because of any incident,” researchers wrote in the report.

Other recent discloses have found models posting user-submitted pictures to third-party hosting sites, as well as an apparent attack on the databases of Australia’s national health service.

Still, it’s likely the new disclosures are just a small portion of the incidents that have taken place so far (we’ve reached out to OpenAI and asked). Axios is reporting major labs have seen as many as 10,000 incidents in which models went beyond evaluator instructions.

OpenAI CEO Sam Altman has implied as much, saying in a post on X on Friday that the company is still sifting through “petabytes of agent activity logs, and working with impacted organizations,” and disclosing incidents “based on severity.” If there’s any consolation in that to be found, it is that Altman says that the Hugging Face incident is still the most severe one OpenAI has found has found. The upshot is, the recent string of rogue agent incidents may be a persistent feature of contemporary frontier research.

Meta announced Monday that it’s launching “Meta Enterprise Platform,” a new initiative aimed at expanding the company’s AI offerings to businesses and corporate customers. The social media giant hired Chirantan “CJ” Desai, the CEO of database software giant MongoDB, to lead the new initiative.

The launch of the new business builds on the momentum of Muse, Meta’s personal AI assistant launched earlier this month that can perform tasks for users such as sending emails and booking travel.

Meta says it will focus on bringing its full technology stack, including Muse, Meta Business Agent, Muse API, Muse Code, and more to businesses and developers.

“Over the coming years, AI will fundamentally redefine how organizations of all sizes innovate, grow, serve customers, and run business operations,” Desai said in a statement. “Meta has a unique role to play because it is bringing together advanced models and leading agents with a proven track record of helping millions of advertisers and hundreds of millions of businesses scale. Meta Enterprise Platform will focus on turning its AI stack into products and services that companies can deploy for their own businesses.”

The move could help Meta see a return on all the money it’s pouring into AI.

MongoDB’s shares dropped by more than 17% on the news of its CEO’s sudden departure. The database maker said it appointed Dev Ittycheria as interim chief executive, who previously served in the role, while the board searches for Desai’s permanent replacement.

Pre-orders for the new iPhone Duo don’t start until later in October, but the device may already have its first killer app. Or at least its most whimsical one! Indie developer Vidit Bhargava has been showing off an app for the first Apple foldable that replicates the experience of using a classic Walkman.

The app, called Duo-Man, takes clever advantage of the Duo’s two screens — a foldable, 7.6-inch inner display when it’s open, and the 5.4-inch outer display when it’s shut closed. The app imitates how walkmans used to function: you can open up the Duo to pick your music and “insert” your cassette tape, then close the device shut to start listening.

Of course, you’re not actually using cassette tapes and players with physical buttons, but Bhargava says he tried to make the experience feel as authentic as possible by recording the button clicks and the static noise from a real Walkman to incorporate into the app, he said.

“I have been a long-term Walkman enthusiast — still own and listen to it occasionally — and have repurposed them in the past,” the developer told TechCrunch. “So, when I was looking for ideas for the hackathon, I was looking for physical objects that have a hinge. I noticed how the Walkman’s door needed to be opened to insert a cassette, and I thought it would be cool to replicate on the Duo,” he said.

While foldable smartphones have been around for years, the iOS developer community tends to be particularly creative, partly thanks to Apple’s App Store ecosystem and its robust tools that let developers build, host, market, and profit from their software. It’s why you often see developers building their apps for iOS first before considering an expansion to Android.

However, the app also taps the broader “retro” trend in tech, as younger consumers seek out classic, physical tech like digital cameras, flip phones, CDs, landlines, and yes, even cassette tapes.

Bhargava initially built the app and demoed it at the Bitrig hackathon this weekend, where it won second place. Bitrig offers software that lets developers use AI to build native apps using Apple’s Swift programming language.

After the developer posted about the Duo-Man on X, it quickly attracted attention, topping 1.8 million views and nearly 10,000 likes. That attention has inspired Bhargava to take what was initially just a fun hackathon project and turn it into a real app.

When it arrives, the app will support Apple Music and local files, and will even include a feature that would let you use your Apple Pencil to “roll” the cassette loop, we’re told. On X, Bhargava also said he plans to make the fast-forward and rewind experiences similar to the original Walkman. Spotify support may be added further down the road.

This is not Bhargava’s first app. The indie developer has published a dictionary and vocab builder, LookUp; a movie recommendations app, Movie Buzz; a time zone utility; and others.