This fortnight we look at how an OpenAI agent broke into an Australian government website because it wanted some statistics, and then offered up an apology three months later in an email. We also talk about Google opening up a tool that lets anyone check whether an image, video or voice clip was made by a machine. The edition throws the spotlight on a Goldman Sachs report suggesting that new graduates will be managing AI agents from their first day, which raises an obvious question about the people who currently manage the graduates. And a young researcher whose former employer sued him for sabotaging its AI now runs a lab valued at $200 million.
1. OpenAI's Agent Hacked Medicare, and the Warning Went to a Public Inbox
In June, during an internal test, an OpenAI model was asked to find out how much the Australian government spends on medicine. It couldn't find the figures through the normal channels, so it read the source code of the Medicare Statistics service and found a way to make the server follow its instructions without an account or a password. OpenAI spotted this in August. It told the government on 10 September, by emailing a general Services Australia inbox that someone checks once a day. The email ran to five paragraphs, recommended the government look into the vulnerability, and was signed "Best, OpenAI Security Team." According to the Guardian, OpenAI used AI to help write it.
No personal data was taken. Prime Minister Anthony Albanese called the breach "obviously unacceptable", and his deputy, Richard Marles, said the agent had "scaled the fence". Sam Altman declined a summons from the Senate, citing short notice, and sent his chief strategy officer, Jason Kwon, to a hearing in Sydney on 6 October. Kwon opened with an apology and said it "should not have happened". Canberra has set up a task force to work out whether anyone can be prosecuted.
What this means for you:
Give an AI agent a goal and leave it to choose the method, and it will try methods nobody signed off on. That holds whether the agent is connected to a government database or your supplier portal.
2. Goldman Says Juniors Will Manage AI From Day One
According to Goldman Sachs' Kevin Sneader, new finance hires will supervise AI agents from the start of their careers. Oversight of routine work, which has traditionally been the job of middle managers, would move down to the most junior people on the team. It fits a pattern at Goldman, where chief information officer Marco Argenti has described the bank's youngest staff as its "AI natives" and said agents will be organised in layers much as people are.
What this means for you:
If graduates manage the agents and the agents do the routine work, the role of the person who used to check that work changes first. Firms that rely on a layer of supervisors to review output should look at what that layer will actually do in two years.
3. Google Opens Its AI Detector to Everyone
On 7 October, Google made its SynthID Detector free for anyone in the world to use, in English. Upload an image, video or audio clip and it tells you whether the file was made with AI from Google or one of its partners: OpenAI, Nvidia and Kakao so far, with Apple joining soon. Google has been embedding an invisible watermark in AI content since 2023 and says it has marked more than 180 billion images and videos, along with 240,000 years of audio. The checks already built into Search, the Gemini app and Chrome handle over a million requests a day.
The catch is that it only recognises content from companies that apply the watermark. A fake made with a tool outside that group will come back clean.
What this means for you:
Checking whether a suspicious image, video or voice note is synthetic now takes an upload rather than a specialist. A clean result can tell you that the file didn't come from a participating company, which is not the same as telling you it's real.
4. Samsung Expects Its Profit to Rise Ninefold on AI Chip Demand
Samsung Electronics expects operating profit for the three months to September to reach 107.4 trillion won, about $80 billion, roughly nine times what it made in the same period last year. It would be the company's fourth record quarter in a row. The surge comes from memory chips used in AI data centres, a market Samsung shares with South Korean rival SK Hynix and Micron in the US, all of which supply companies such as Nvidia. Full results are due at the end of October, though Korean companies' own previews are generally trusted more than analyst estimates because they draw on internal figures.
Demand has outrun supply. Google, Amazon and Meta have committed more than $650 billion to AI projects this year, South Korea announced plans in June for at least $880 billion of chip manufacturing projects led by Samsung and SK Hynix, and Japan, China and Taiwan are building new plants too. In the meantime the shortage has let Samsung and others raise prices, and smartphones and computers have become more expensive as a result.
What this means for you:
The AI boom is now raising the price of ordinary hardware. Any budget for laptops, phones or servers over the next year should assume higher prices, and buying ahead of need may cost less than waiting.
5. Sued by ByteDance, Now Running a $200 Million AI Lab
Tian Keyu used to be an intern at ByteDance, the company behind TikTok. In 2024 ByteDance sued him for 8.02 million yuan, accusing him of altering code and disrupting its model training. The court accepted the case and no ruling has been made public. That same year, Tian was lead author on the paper that won the top award at NeurIPS, the field's most prestigious conference, for a new way of generating images. Bloomberg reported on 7 October that his new lab is valued at about $200 million and is going after the same ground as Fei-Fei Li's World Labs: "world models", which are AI systems built to understand three-dimensional physical space. World Labs raised $1 billion in February.
What this means for you:
Investors are backing proven research talent even with a lawsuit attached. The handful of people who can build these systems are scarce enough that a pending case from a former employer hasn't put the money off.
About Us
At AI-First Mindset, we help leaders bridge the gap between knowing AI and using it. Our work includes hands-on workshops and operating frameworks designed for companies that want to embed AI into workflows without relying on large institutional vendors.
To explore how we can help, contact us at aifirstmindset.ai.