AI Agents Breach 100 Firms, OpenAI Races Toward $1.4 Trillion

Full transcript

Alex: Welcome to Daily Tech News for Tuesday, October twentieth, twenty twenty-six. I am Alex, and today the A I agents are on both sides of the security story, because a hacker reportedly let agents break into more than one hundred company websites, while Microsoft rolled out new tools designed to shut risky agents down.

Jordan: I am Jordan, and we also have a money race, because OpenAI is reportedly rushing a huge funding round ahead of Anthropic's expected public listing. We will also cover a one point five billion dollar bet on an Nvidia challenger, a federal order removing DeepSeek from civilian systems, and Apple's tease of new Macs.

Alex: Let us start with the breach campaign, because its economics should worry people. Reuters reports that a Chinese speaking hacker broke into websites belonging to more than one hundred companies after assigning much of the work to A I agents.

Jordan: The key detail is how little human direction that took. The attacker could hand the system a goal and let the agents carry out much of the attack sequence on their own, and the whole campaign reportedly cost only a few thousand dollars.

Alex: It is worth being precise about what this is not. This is not a model escaping a lab on its own. This is a human attacker deliberately pointing agent tools at targets, and finding that one person can now attempt far more intrusions than a single hacker reasonably could before. For defenders, that means patient, multi step attacks are now available to almost anyone, and volume will rise.

Jordan: Here is the other side of that coin, because on Monday Microsoft introduced new controls meant to police the agents that companies themselves deploy. The features are part of Microsoft three sixty five E seven, the company's top tier workplace bundle, and of its Project Perception security system.

Alex: The idea is to treat every agent like a new employee with system access. Each agent gets a continuously updated risk assessment, based on what data it can reach and whether its actions are drifting from its assigned task.

Jordan: And when that risk rises, the response is graduated. The system can take away an agent's ability to make changes while still letting it read data, require a person to approve a high stakes step, or shut the agent down entirely.

Alex: Microsoft is also adding Guardian Agents that watch other agents and recommend or take containment steps. Detection on employee computers is expanding from twenty five types of agents to sixty, including Claude Code, Codex, and GitHub Copilot.

Jordan: The philosophy came from chief executive Satya Nadella, who argued this month that controls on A I models need to live outside the models themselves. It is also a product, though, because the E seven bundle costs ninety nine dollars per user per month, and the controls are in private preview ahead of broader availability around Microsoft's Ignite conference in mid November.

Alex: From agent security to the money, because Reuters reported Monday that OpenAI is racing to lock in a funding round of about thirty billion dollars, at a valuation topping one point four trillion dollars.

Jordan: That deserves a pause, because OpenAI was last valued at eight hundred fifty two billion dollars in March, so this would be a jump of more than sixty percent in seven months. Backers including SoftBank and Abu Dhabi based M G X are in talks to take part, although nothing has been finalised, and OpenAI declined to comment.

Alex: So why the hurry? The timing is about Anthropic, which is expected to make its stock market debut in mid November. Both companies court many of the same sovereign wealth funds, and once Anthropic trades publicly, investors will have a market tested price to compare OpenAI against. OpenAI would rather set its own price first.

Jordan: Reuters also reports that OpenAI is telling investors its annualised revenue has reached fifty billion dollars and could reach seventy billion by the end of the year, and listeners should hear those as the company's own figures, offered while it is raising money. The honest caveat is that this round has not closed, so landing that price is a real test of the market's appetite.

Alex: Speaking of infrastructure money, Tenstorrent announced on Monday that it has closed a one point five billion dollar funding round at a six point two billion dollar post money valuation, more than double its value since December twenty twenty-four.

Jordan: The round was led by Trace Capital and Innovation X, with backers including Fidelity, Morgan Stanley, T. Rowe Price, and A R K Invest. Tenstorrent says the money will help it deploy its Galaxy servers at scale and advance the chip designs it licenses to customers building their own silicon.

Alex: The pitch is openness. Tenstorrent builds on risk five, an open chip architecture anyone can use without licence fees, and its software is open source. Its chief executive is Jim Keller, a famed chip designer with senior roles at Apple and A M D and a stint leading Tesla's self driving chip work.

Jordan: The sceptical view is that Nvidia's moat is not just silicon, it is the software ecosystem around it. But a cheque this size gives customers who want a second source for A I compute a much better funded option to test.

Alex: Next, a policy story that became public this week, even though the decision is older. The U S Department of Homeland Security has ordered DeepSeek products removed from federal civilian agency systems, and barred agencies from buying them in the future.

Jordan: What makes the order broader than people might expect is that it also covers third party products and services developed using DeepSeek implementations. So it is not just the DeepSeek app, it is any vendor tool built on those models.

Alex: The department said the models present multiple threat vectors, and that using them could result in sensitive data being transferred to China. The order was issued back in July, under a twenty eighteen supply chain security law, and the public notice only appeared this week.

Jordan: Two limits are worth stating clearly. The order covers civilian agencies, it does not cover the Department of Defense, and it is not a ban on private businesses or consumers. What it does is put vendors on notice that federal buyers may need to know which underlying model sits inside the products they purchase.

Alex: Finally, a lighter closer. Apple's marketing chief Greg Joswiak teased new products on Monday, writing that he cannot wait to, quote, pro and tell, next week. Apple has invited media to an experience in Los Angeles on Wednesday, October twenty eighth.

Jordan: We should separate what is confirmed from what is rumoured. The tease and the event are confirmed, but Apple has named no products. Reporting from Bloomberg and others expects a redesigned MacBook Pro with a touchscreen O L E D display, a first for the Mac, alongside an entry level MacBook Pro, an iMac running the new em six chip, and an iPad mini with an O L E D display, possibly on sale Friday, November sixth.

Alex: So let us pull the threads together. Today's stories share one theme, and it is authority. A hacker handed authority to agents and attacked at scale, Microsoft is selling tools that can revoke an agent's authority mid task, and Washington just revoked a model family's authority to sit inside civilian systems.

Jordan: And the money sets the stakes, with OpenAI chasing a valuation above one point four trillion dollars and Tenstorrent raising one point five billion to prove the chip market still has room for challengers. Thanks for listening, and we will see you tomorrow.