In July, OpenAI admitted that one of its agents tasked with completing a cybersecurity experiment broke out of containment and hacked AI dataset platform Hugging Face. That incident, which got a full accounting from OpenAI yesterday, was the first publicly reported case where an LLM went rogue and autonomously hacked a third party. 

Since then, that unprecedented sci-fi-esque event turned out to be far less rare than anyone would hope for. 

According to a satirical website called Felony Bench (for benchmark), which tallies these incidents, there have been 17 incidents in total. It’s important to remember that criminal law experts are not entirely sure whether the AI companies that made the LLMs that did the hacking can be prosecuted, nor whether the victims can sue them. But we are likely going to get an answer to those questions soon.  

Anthropic and OpenAI’s models lead the race with eight incidents each, and Meta trails behind with one, according to the site. At this point, it has become clear that AI safety tests are becoming safety risks themselves. And some AI companies and workers themselves have recognized those risks in the “Pacing The Frontier” open letter, which called for developing AI capabilities responsibly. 

We decided it would be a good time to recap all these incidents chronologically.

nternet access. From there, several agents worked together to target and hack Hugging Face thinking they could find the solution to the challenge there. OpenAI only found out after Hugging Face disclosed it had been a victim of a fully autonomous attack. Whoops. 

Anthropic discloses it hacked three companies

OpenAI’s disclosure piqued the curiosity of Anthropic, who wondered: could this have happened to us too? Turns out, the answer was yes. Three times yes. The frontier lab discovered that its own models breached three different and still unnamed companies, with the earlier incident dating back to April—more than three months before the company discovered it. Anthropic partially blamed Irregular, a startup that runs AI cyber evaluations. Whoops.

OpenAI finds out that, actually, Hugging Face wasn’t the only victim

Once OpenAI started investigating the Hugging Face breach, it found out that the agents that hacked Hugging Face also broke into four accounts and four different companies, as Reuters first reported. Modal, an AI inference startup, was one of the victims. Whoops.

Irregular realizes an OpenAI model hacked a company

In late July, Irregular told OpenAI that one of its models that was participating in a Capture-the-Flag competition — essentially a cybersecurity game where players hack systems designed specifically for the competition — escaped the game, connected to the internet, and hacked a real company. The reason? Irregular had given one of the fictional targets the same name of a real company. Whoops. 

UK’s AI Security Institute tries to hacks “real people and organizations”

Also in late July, the UK government’s AI Security institute, a public body tasked with researching the safety and risks of AI technologies, disclosed that it detected several incidents involving both OpenAI and Anthropic models that while running “routine” evaluations targeted “real people and organisations.” In these cases, AISI had given the models internet access. Whoops. The good news is that the agency actually detected as they happened, rather than weeks later like in other incidents. 

In early August, Meta became the last company to disclose an incident involving one of its LLMs, which hacked “a third-party” service. Meta blamed the incident on a misconfiguration by Irregular, which was running a cybersecurity valuation for the tech giant that was supposed to not have internet access. Whoops. 

Claude agent hacks gym’s software to book a class

An Australian man asked an Anthropic AI agent to help him book a gym class, which he was on a waiting list for. “I was just sitting on the couch thinking, ‘Gee, this is a chore,’” the man told ABC Australia. In its attempt to comply with the request, the agent found a vulnerability in the gym’s booking software, exploited it, and kicked out people who were ahead of the man on the waitlist. The man tried to undo the damage, asking the agent to undo its actions. The agent replied: “Bad news — I can’t add them back.” Whoops.

Hardware companies have realized that note-taking is one of the easiest AI use cases to build for, and consequently have been busy shoving mics into everything from pendants and rings to credit-card-sized pucks and wristbands. Still, despite the variety of devices, only a few companies have been able to stand out.

One of these companies is Plaud, which has over 2.5 million people using its hardware or software. It is now launching a note-taker in a new form factor: earphones. Called Plaud One, these adopt the simple bare-bones style of Apple’s Airpods, and can record calls, while their case can be used to record in-person conversations or take notes.

Plaud isn’t the first to this form factor. Its rivals Viaim and Anker already have similar note-taking earbuds in the market.

But the company is betting on its software and AI chops to set these apart. Plaud said that once the earbuds transcribe your conversation, its AI agent — which can connect to tools like Gmail, Google Calendar, Notion and Slack — can do things like writing follow-up emails and creating documents or presentations based on the contents of the conversation.

By itself, that isn’t a unique pitch. Software note-takers like Granola, Fathom and Read AI have been offering similar workflow automation for a while now, transcribing calls and conversations, and allowing you to subsequently get an AI agent to do various tasks.

The Plaud One takes that a step further, though, with its case, which supports an eSIM card. That feature, the company says, allows users to instruct Plaud’s AI agents remotely without using their phone or computer.

The Plaud One has 12mm drivers and features active noise cancellation. The company says the case can capture voices clearly within a range of five meters, and the earphones can record six hours of in-person meetings and three hours of calls on a single charge. Combined with the case, users can record up to 25 hours of conversations on a single charge.

Buyers get 300 minutes of free transcription per month, after which they will need to buy additional plans, which start from $8.33 per month (paid annually).

That said, this launch seems to be more of a trial run. The Plaud One will be limited in quantity, as the startup likely wants to gauge demand for such AI-focused earphones before it commits to the format.

Plaud is also working on building agentic capabilities, which is expected to arrive as part of an update in the coming months. The company says it will introduce a credit-based system for different tasks, and buyers of the Plaud One will get $200 in credits for these tasks.

The company has had a good run so far, claiming in June that it had reached annual run-rate revenue of $100 million. However, competition in the market is intense, with new hardware note-takers releasing every week. Its rival, YC-backed Pocket, has also reached that milestone, and tons of other companies like Anker, Comu, Viaim, and software startup Genspark are building in the space.

Plaud’s CEO Nathan Xu holds that there should be an interface to talk to AI to invoke it from anywhere, and earbuds meet that need.

“[For AI] we need to have a wearable and an always-ready interface, one that people love to wear, is socially acceptable and private by design,” he said in a briefing.

Still, the company may find it challenging to draw people away from using their AirPods, Google Pixel Buds, Samsung Galaxy Buds, or wireless earbuds from any number of audio companies, which can pretty much do the same job when paired with any note-taking app.

But Xu says Plaud is not necessarily competing against these companies, but positioning its meeting-recording tech to adopt a popular wearable format.

Users can pre-order the device now for $249, and it will ship in the fourth quarter of 2026.

It seems there is no escaping the ad industry.

OpenAI said on Thursday that it will start showing ads on its ChatGPT Free and Go subscription tiers in India. The move comes after the company changed its terms of service earlier this month to indicate that it will show ads to users while they use the AI assistant.

OpenAI said in February that it has more than 100 million weekly active ChatGPT users in India, a huge chunk of whom are on the free or the lower-priced Go tiers.

The company said on Wednesday it will start by showing ads for 50 brands, and that it has partnered with agencies WPP and Omnicom for this. OpenAI will also launch an ad manager next month that will let marketers create campaigns for their companies, provided their campaigns have a daily minimum budget of ₹725 ($7.60).

“With ChatGPT Ads, businesses of every size can introduce themselves at relevant, high-context moments when decisions are beginning to take shape,” Dave Dugan, head of global ads solutions at OpenAI, said in a statement.

OpenAI has worked hard to cultivate its user base in India, launching a sub-$5 plan for ChatGPT Go in August 2025, and even running a limited promo that made that tier free for a full year. It also heavily advertised its products during sports tournaments, like the Women’s Premier League (WPL) and the Indian Premier League (IPL) cricket leagues. More recently, it hired Uber’s India head to lead its expansion efforts in the country.

The AI lab started showing ads to U.S. users in February, and expanded the program to serve ads in Europe earlier this month.

Ahead of a potential IPO, expected this year or next, OpenAI has been trying to ramp up and cement its revenue sources. According to The Wall Street Journal, the AI lab recorded revenue of $6.7 billion in the second quarter ended June 2026, up from $5.7 billion the previous quarter.

Last November, The Information reported that OpenAI was aiming to reach 220 million paying subscribers by 2030. The report said that at that time, 35 million users paid for its Plus and Pro plans.