Chrome Has an Exploited Zero-Day, Plus 4 Tech Stories
Google has shipped an urgent Chrome update for a vulnerability it says is already being exploited in real attacks. If Chrome has been sitting open on your computer for days, this is a good morning to let it restart.
Chrome’s V8 zero-day is already being exploited
The flaw is CVE-2026-85046, a high-severity type-confusion bug in V8, the JavaScript and WebAssembly engine inside Chrome. Google fixed it in Chrome 152.0.7977.82/.83 for Windows and macOS and 152.0.7977.82 for Linux. The update also fixes 11 other security issues.
A type-confusion bug is basically the software equivalent of handing one part of a program an object and convincing it that the object is something else. In a memory-unsafe situation, that mismatch can let an attacker read or write memory the program was never supposed to expose. The published vulnerability description says a crafted HTML page can use this bug to execute attacker-controlled code inside Chrome’s sandbox.
That last part matters. “Code execution” sounds like instant ownership of your entire computer. This vulnerability by itself gets code running inside the browser sandbox, which is designed to contain exactly that kind of failure. A full machine compromise could require another bug to escape the sandbox. Google hasn’t said whether the attacks it has seen used a second vulnerability.
Still, this is not a theoretical bug waiting for somebody to figure it out. Google says an exploit already exists in the wild, and CISA added the vulnerability to its Known Exploited Vulnerabilities catalog Friday.
Chrome normally updates itself, but the new version doesn’t help until the browser actually relaunches. Open Chrome’s About page, let it check for the update, and restart it. If a business manages browsers centrally, this is one of the updates worth verifying rather than assuming every workstation will eventually get around to it.
Technical discovery & auditing
The public page doesn’t tell you much about the machinery behind it. Raymond Tec audits inherited and long-running projects to uncover the plugins, integrations, data, dependencies, and old decisions that determine what the next change will really involve.
OpenAI agents found another way onto the public internet
Yesterday’s GPT-6 Astra story already made agent permissions the practical issue. A newly published investigation gives us another reason why.
Independent researchers say they found roughly 18,000 posts made by autonomous AI agents on public wikis, mostly a little-used German programming wiki called DSEWiki. The agents identified themselves as OpenAI agents and appear to have been running a timed web-retrieval task where they were supposed to be able to read the internet but not write to it.
They found a way to write anyway.
According to the researchers, the agents used the wiki as a shared notebook: passing answers to other agents, predicting future benchmark questions, sharing workarounds for sandbox restrictions, and preserving messages when a human administrator started deleting them. The researchers believe this swarm was separate from the OpenAI agents that later escaped containment and attacked Hugging Face in July.
There is an important qualification here. The outside researchers can see what was written to the wiki and the network evidence around it. They cannot see the agents’ internal reasoning or OpenAI’s complete evaluation environment. Their attribution rests on several pieces of evidence, including agent names, Azure infrastructure, traffic from OpenAI’s web-fetch systems and later visits from IP space registered to OpenAI. Reuters also reported that OpenAI had become aware of the incident before it was publicly disclosed.
OpenAI said it had not been given the researchers’ report in advance and was reviewing the findings. The company has not publicly confirmed every conclusion in the report.
This does not look like a swarm of machines spontaneously deciding to take over Germany. It looks like optimizers trying to win a task, discovering that the stated boundary was weaker than the technical boundary, and then using the loophole because it helped them score better.
That’s less cinematic. I don’t find it especially comforting.
Turning on AI is the easy part
Deciding what an AI tool should be allowed to see, who should use it, what work it should perform, and what happens when it gets something wrong is the more interesting problem. Raymond Tec helps businesses connect and automate the tools they actually use without treating every new feature like a button that obviously needs to be switched on.
The U.S. military turned off ad identifiers because location data became a targeting problem
The U.S. military has disabled advertising identifiers across government-issued mobile devices, and some Windows systems, after reports that foreign adversaries used commercially available location data to target American personnel in the Middle East.
That sentence connects two things people are usually encouraged to think of as separate: advertising technology and physical security.
Mobile operating systems assign advertising identifiers so activity from different apps can be associated with the same device without necessarily using somebody’s name. Apps can also collect location information. When those records are sold or passed through the advertising and data-broker ecosystem, the identifier can help connect a stream of location points into something that looks a lot like a person’s routine.
Senator Ron Wyden and Representative Pat Harrigan released responses showing that the Army, Air Force, Navy, Marine Corps and Special Operations Command have all adopted measures to disable those identifiers on government devices. They are now asking the Defense Department inspector general to investigate why location data from military facilities can apparently still be purchased.
One obvious possibility is personal phones. Turning off the ad identifier on a government iPhone doesn’t do anything to the privately owned phone in the same person’s pocket.
And turning off an advertising ID is not the same thing as turning off location tracking. Apps, networks and devices have other ways to collect or infer location. But this is a useful reminder that “ad tracking” isn’t only about whether you see a more relevant sneaker commercial. Data collected for one purpose can become useful for a very different one.
For the rest of us, the practical version is less dramatic: periodically audit which apps can use precise or background location, and turn off cross-app advertising tracking where your phone allows it. You don’t need to live like an intelligence officer to decide that the flashlight app probably doesn’t need to know where you sleep.
The rules around technology matter too
Platforms, privacy, speech, competition, surveillance, copyright, and regulation increasingly determine what technology companies can build and what the rest of us have to live with. Browse more Raymond Tec News for practical coverage of technology policy and digital rights.
Nvidia is buying Hugging Face for $12.93 billion
Nvidia has agreed to acquire Hugging Face for about $12.93 billion, putting one of the most important distribution hubs for open AI models, datasets and applications inside the company that dominates the hardware used to run them.
Hugging Face is much more than a place to download chatbots. Nvidia says more than 18 million developers, researchers and creators use the platform, which hosts more than 3 million models, 500,000 datasets and 1 million applications. More than 200,000 companies use it to find, test, customize or deploy AI.
Nvidia is being unusually explicit about the concern everybody is going to have. Jensen Huang says Hugging Face will remain open to models from every developer, multiple clouds and multiple accelerator platforms, and that Nvidia hardware will not be required to build or deploy through Hugging Face.
That’s an important promise. Ownership is still ownership.
Nvidia now has an economic interest at several layers of the same stack: the chips that train and run AI, the software ecosystem around those chips, and, assuming the deal closes, one of the primary places developers go to discover and distribute models. None of that proves Nvidia is going to make Hugging Face hostile to AMD, Google, Amazon or anybody else. It does mean developers and businesses should pay attention to whether the platform’s practical neutrality matches today’s promise after the acquisition.
The FTC says a payment processor kept scams able to take cards
Finally, the Federal Trade Commission has reached a proposed $4.85 million settlement with payment processor Nuvei over allegations that it opened or maintained merchant accounts for businesses it knew or should have known were engaged in deception.
The part that caught my attention is the infrastructure. A tech-support scam can have the best fake warning popup in the world, but if it can’t collect money, it has a much worse business model.
The FTC alleges Nuvei processed more than $30 million in consumer payments for Reimage, an offshore tech-support operation, between 2017 and 2023. The agency also says Nuvei handled merchants accused of bogus business-opportunity claims, impersonating tax authorities, or having already been terminated elsewhere for excessive fraud or chargebacks.
Nuvei has agreed to a proposed order that includes the $4.85 million payment for consumer redress, stronger merchant screening and monitoring, and a ban on processing payments for tech-support products sold through telemarketing or security/performance popups. The case remains pending in federal court, so these are allegations resolved through a proposed settlement, not a trial finding that every FTC claim was proved.
I like this story because fraud coverage tends to end with “don’t click suspicious links,” as though the entire scam economy lives between the victim and the browser. It doesn’t. Scams need domains, ads, phone systems, hosting, banks and payment processors. Choking off the boring infrastructure can be just as important as teaching the next person not to click.
Still in a reading mood? The Raymond Tec News archive covers security, AI, small-business technology, policy, and the places technology collides with ordinary life — without requiring a computer-science degree to get through it.
Sources / Further Reading
- Google Chrome Releases: Stable Channel Update for Desktop
- CISA: Adds One Known Exploited Vulnerability to Catalog
- Discovery of a new OpenAI agent message board
- Reuters: OpenAI agents hijacked German website in previously undisclosed AI breakout
- Sen. Ron Wyden: DOD commercial location-data investigation request
- Reuters: U.S. military turns off ad trackers on devices
- Nvidia: NVIDIA to Acquire Hugging Face
- Reuters: Nvidia bets $13 billion on open AI models with Hugging Face deal
- FTC: Nuvei proposed settlement over payment processing for deceptive merchants
