Ant and Anthropic
Earlier this year, Anthropic’s new frontier AI model made headlines. Mythos was impressive enough as a standalone model, but it was the cybersecurity implications that made the world skip a single heartbeat, and I’m not sure we’ve seen even that much of a pause so far during the rapid rise of AI.
I have read a lot of breathless headlines like these over the years, but usually the AI story goes away before the national security story does. Mythos was the first time this wasn’t the case.
Suddenly, it seemed like every system in the world was now vulnerable to hacking. Unlike previous AI models that had become good at exploiting specific vulnerabilities, this model went about discovering openings relentlessly—and cleverly.
Exactly how big of a deal this was became clear when Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent called an urgent meeting with the CEOs of Bank of America, Morgan Stanley, Goldman Sachs, Wells Fargo, and Citigroup to Washington. Here in this closed-door meeting, the CEOs were told directly of the imminent danger now posed to their existing security systems.
Suddenly, worried readers thought, your 401(k) might not be so secure any more. Yikes!
It was soon after that the government took the unprecedented step of effectively banning the model from all use, at least temporarily. If the urgent meeting between the government and the banks didn’t get everyone's attention, this certainly did. No other AI model had been pulled from the shelves like this. Paying US customers who had previously had access were now denied.
The government hadn’t technically banned Mythos, but merely suspended foreign-national access to the model. However, since Anthropic claimed it couldn’t be sure it was complying with this user-by-user approach, the company simply disabled Mythos 5 and Fable 5 for all customers.
Many observers thought that Anthropic was paying an unfair price for its apparently principled stance on what their tools should be used for. CEO Dario Amodei had been a vocal critic of AI for surveillance, and Anthropic had consistently refused to remove safeguards on their models at the request of the government. This seemed to contrast other AI leaders who were more willing to comply with these types of requests.
Anthropic had already been designated a supply-chain risk, and now a model had been effectively banned from public use.
There is also the matter of the government ordering all federal agencies to stop using Anthropic. Many argue that this was in response to the politics of Amodei and little else, so it looks like a punitive action by an overreaching government.
This is all eerily familiar to me.
Jack Ma, the founder of Alibaba, was on top of the world just a few years ago. Alibaba was a bit like the Amazon of China: tech heavy, automation centered, and with a strategically important cloud computing business. Ant Group was a finance spinoff from Alibaba, with Alipay as its crown jewel. It was now worth several hundred billion dollars in its own right.
On October 24, 2020, Ma was a keynote speaker at a major finance summit in Shanghai. He was scathing and unflinching in his criticism of what he saw as an old order, inhibiting innovation and growth with a “pawnshop mentality.”
Now, I wasn’t terribly surprised when Dario Amodei began criticizing the US government loudly. This was very much not the case when I first heard of Jack Ma voicing his discontent with the direction of the Chinese Communist Party.
It didn’t take long for them to act.
Less than two weeks later, Ant’s planned IPO—planned as one of the biggest IPOs in history—was halted by Chinese regulators. The valuation of Alibaba crashed rapidly.
Then, Jack Ma disappeared.
For months, there were no public reports of Ma anywhere. Rumor mills churned overnight, speculating that he had been killed in order to silence him, with online jokes reminiscent of Russia’s tall building problem.
Eventually, Ma resurfaced with a much humbler outlook, and a much poorer portfolio. Many thought he was lucky to walk away with his life.




ANTHROPIC COPIED MY BOOK!
They put my animation history "America 'Toons In" in their AI-training system without the permission of either me or my publisher. I put my name into the class-action lawsuit related to the training against them, and I am supposed to be financially compensated, but the funds haven't arrived yet...
Mythos timing was good for me. I was teaching the TCP/IP stack to my HS kids when that news hit and one of the vulnerabilities Anthropic cited in their ‘red team’ brief on it was a 30 year old Unix bug in the TCP/IP stack. The same BSD stack i learned to code on. This was a teaching moment I couldn’t skip: connecting today with the reality of the plumbing that runs the entire internet.
Did I write about it you ask? Of course I did: https://newsletter.wirepine.com/p/when-your-brain-hurts-from-thinking
In another China twist, I believe Anthropic recently accused China of sabotaging Mythos (or was it Fable?) by trying to bleed it with millions of fake agents to train their models. Haha Anthropic - jokes on you - your model that breaks the internet gets hacked.