Apple Beats, Anthropic Warns on Self-Building AI, cPGuard Hit

Full transcript

Alex: Good morning, and welcome to Daily Tech News. Today is Tuesday, and we have three stories that fit together better than they first appear. Apple just delivered a record quarter under its new chief executive, Anthropic just warned that A I may soon start improving itself faster than we can evaluate it, and a security tool meant to protect servers was just turned into a delivery vehicle for an attack.

Jordan: That is quite a lineup, and the thread running through all three is control. Let us start with Apple, because the numbers are in now.

Alex: Apple reported its fiscal fourth quarter results on Monday evening, and they beat expectations. Total revenue came in at one hundred fourteen point five billion dollars, up twelve percent from a year ago. Profit was thirty point seven billion dollars, with earnings of two dollars and ten cents per share. That was the first earnings report under chief executive John Ternus, who took over in September.

Jordan: The star was the iPhone, and this is the interesting part. IPhone revenue hit fifty seven point five eight billion dollars, up seventeen point five percent, and that is a record. It did that with only the pricier Pro models in the latest refresh, because Apple did not release a standard model this time around.

Alex: Right, and the foldable iPhone Duo contributed nothing to this quarter at all, because its sales fell outside the window. Mac revenue also beat expectations, at ten point seven nine billion dollars, up about twenty four percent. Services hit thirty one point five seven billion dollars, essentially in line with expectations. The iPad was the weak spot, down about ten percent.

Jordan: So why did the stock wobble after hours if the quarter beat? That is the margin story.

Alex: Exactly. Gross margin was forty seven point nine percent, better than expected, but Apple guided to forty five to forty six percent for the holiday quarter, below what analysts wanted. The company pointed to rising memory chip prices and investment in its revamped Siri. On the call, Ternus said memory prices rose through the September quarter, and price increases on iPhones, Macs, and iPads helped offset the pressure. He also confirmed the iPhone Duo remains supply constrained, with no timeline for when supply and demand will balance.

Jordan: Still, the holiday guidance is striking. Apple expects December quarter revenue to grow seven to ten percent, which would make it the biggest quarter in the company's history, even with constrained supply. The December quarter will be the first to include Duo sales. So the question for the holidays is not demand. It is whether Apple can build enough of the things people want.

Alex: And that brings us to Anthropic, because there the constraint is not hardware. It is evaluation, and the company just said so in its own filing.

Jordan: On Monday, Anthropic amended its filing ahead of its planned initial public offering, and the language got a lot of attention. The company warned that it may soon cross its own threshold for dramatically accelerated A I assisted research, and that it might not notice the shift until after it happened.

Alex: Let us unpack that, because the phrase doing the work here is recursive self improvement. It means A I getting good enough at building A I that progress starts speeding up on its own. Anthropic is not saying that has happened. It is saying its internal forecasts suggest it could happen imminently, and that its ability to evaluate the models might not keep pace.

Jordan: The timing matters, because this filing landed the same day as a remarkable research result from M I T. Researchers asked an OpenAI model called Astra to edit the weights of a much smaller model directly, from plain language goals, with no training run at all.

Alex: Astra changed only zero point zero one six percent of the smaller model's weights, and the smaller model's score on a standard math test jumped from seven point nine percent to forty eight point one percent. The researchers call the technique weight space programming, and they describe it as very early work.

Jordan: And we should be honest about the caveat, because other researchers pushed back quickly. Some argued the gain may be closer to removing a reasoning loop than rewriting the model's abilities, and at least one independent rerun suggested a simpler setting matched the gain. So the size of the effect is genuinely disputed, even if the demonstration itself is real.

Alex: What is not disputed is the direction. A frontier model edited another model's insides, in words, and the smaller model got much better at math. Pair that with Anthropic saying self building A I may be near, and you get the question that will define next year. If models start improving models, who watches the improvement, and how fast can that watch keep up.

Jordan: Investors, meanwhile, are applying their own discount. Reports say institutional buyers are now pushing Anthropic's valuation down to between one point five and one point six trillion dollars, from two trillion, citing the broader tech selloff and doubts about A I returns. The amended filing also showed quarterly revenue of about seventeen billion dollars, but growth slowing sharply from the quarter before.

Alex: Which is a good bridge to our security story, because it is also about trust in a system that is supposed to watch over you.

Jordan: This one broke overnight. A Linux hosting security tool called cPGuard had its update channel compromised, and the attacker used that trusted channel to push a malicious script to customer servers.

Alex: According to the vendor, known as OPSSHIELD, the breach began on Monday, and the attacker used compromised credentials to reach its update distribution infrastructure. There was no software bug exploited here, at least as far as the vendor has said so far. The attacker simply got into the distribution system and rode the update mechanism.

Jordan: The malicious script went after the most sensitive files on a server, including password hashes, root database configuration, root shell history, and root secure shell keys. The company says it detected the attack within twelve hours, secured the compromised account, and stopped further distribution, but it also estimates the attacker retrieved thirty two collected files, each corresponding to one customer server.

Alex: The fix is straightforward, and it is urgent if you run this tool. Update to version five point eight eight point twenty five or later, then rotate every credential that could have been exposed. And the lesson is bigger than one vendor. When the update channel itself is compromised, the security tool becomes the attack.

Jordan: Before we close, a quick look at the rest of today. It is election day in the United States, and both OpenAI and Anthropic have set up dedicated teams to watch for election misinformation on their platforms. A M D reports earnings after the bell today. The monthly jobs report lands on Friday, and Apple Intelligence testing in China, powered by Alibaba's Qwen, starts tomorrow in the developer beta.

Alex: So here is the takeaway. Apple proved demand is not the problem, supply is. Anthropic warned that evaluation may soon lag behind the very progress it is meant to evaluate. And cPGuard showed that the most dangerous breach is the one that arrives disguised as a trusted update. Three different stories, one question underneath them all. Who is actually in control of the systems we depend on.

Jordan: We will be watching A M D tonight, the jobs report on Friday, and any further word from Anthropic on that filing. Thanks for listening, and we will see you tomorrow.