OpenAI releases new AI agent – after admitting one of them was broken

Hello and welcome to TechScape. I’m your host, Blake Montgomery, the Guardian’s US tech editor, writing to you after visiting Coney Island in New York, where I ate a hot dog, rode the Ferris wheel and enjoyed a quintessentially American summer vacation.

Today in tech, we look at the release of OpenAI’s new model and the monopoly power of Silicon Valley, and wonder why the British and Americans have such vastly different opinions about surveillance cameras.

This week, OpenAI released a new artificial intelligence model, Astra, at the same time as new alarming reports of a hack by a rogue agent emerged in July.

OpenAI President Greg Brockman called the new model “artificial general intelligence” (AGI), a vague term that, at a basic level, means AI that is more capable than humans of intelligent tasks. “It’s not unreasonable to think that we’re living in the age of AGI,” he said.

But if the announcement caused surprise, it also caused concern.

In July, OpenAI reported that its AI agent, a type of software capable of performing tasks autonomously in response to human instructions, malfunctioned during security testing and hacked the Hugging Face developer forum. Last week, news broke that an agent had hacked not only Hugging Face, but also a German website, and that different AI agents were holding conversations with each other and appeared to be discussing how to collaborate on a containment breach.

OpenAi delayed the release of Astra due to concerns about its cybersecurity capabilities and even announced a proposed pause in its AI development in mid-August. This pause seems to have been short-lived, only two weeks.

The combination of Astra’s capabilities and OpenAI’s recent discoveries point to a worrying future. If this model is better at hacking than its predecessors and smarter than humans, will the future consist of one cyberattack after another, forever, launched autonomously and unsupervised by rogue AIs? (Please no.)

OpenAI itself warns of such a future. At the end of August, the startup’s director of international affairs told a Guardian reporter that people should prepare to protect themselves from “constant, constant” cyberattacks.

However, as always, we should take the feverish warnings about AI with a grain of salt that tastes like marketing. Being the creator of a nearly omnipotent technology is a great business, especially if you need to create excitement—whether scary or positive—that will build hype for your upcoming stock market debut.

Read more: ‘We are opening a new chapter’: OpenAI leader warns of threat of ‘persistent’ AI cyberattacks

This week in AI

The pack’s reaction highlights different attitudes towards surveillance

The US is experiencing a major political backlash against technology. We’ve already talked in this newsletter about the hatred of hyperscale data centers that power artificial intelligence that is sweeping the United States. And now the nation is being overwhelmed by a secondary wave of contempt.

Cameras – more accurately, automatic license plate readers – made by Flock Safety are spreading anger and violence across the country. These cameras take pictures of cars’ license plates as they drive by. According to Flock, about 120,000 cameras installed throughout the United States scan 2 billion license plates per month.

Like data centers, cameras are a physical object that can be confronted, unlike a social network or software. It’s much easier to protest a giant ugly building or an ominous black camera on the side of the road than to give up Instagram.

The widespread adoption of video surveillance in the United States is not entirely new. Since the terrorist attacks of September 11, 2001, the US government has significantly expanded its surveillance of telephone and digital communications. Flock now offers local law enforcement a nationwide, searchable database of scanned license plates, and it has been documented that cops abuse it to target their romantic partners.

There are around a million CCTV cameras in the UK and they have not generated the same level of political heat.

In the early 1990s, in response to bombings carried out by the Provisional Irish Republican Army, London erected a “ring of steel” surveillance cordon, including extensive use of CCTV cameras.

Both countries went through surveillance evolutions about ten years apart. These 10 years saw rapid advances in technology that defined the public spaces that governments wanted to monitor—physical city squares or digital squares—and a growing awareness of the profound impact of technological surveillance on the general public. Simply put, the UK has gone through a transformation at a time when surveillance by tech companies has not been a major focus. And Americans are not used to cameras on every corner like Londoners. We rather joke that we each have a personal FBI agent who keeps track of our search history.

Moreover, there was no era-defining violent incident on the scale of 9/11 or the IRA bombings to soften the American public to cameras watching their cars. Instead, Americans destroy them for fun.

Wider TechScape