Jun 28, 2026 · 19m · big-technology
AI’s True Cyber Risk: Fable, Mythos, China, and The Rest — With Alex Stamos
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Cybersecurity expert Alex Stamos analyzes the U.S. government's abrupt shutdown of Anthropic's frontier AI models, Fable and Mythos, examining the crucial technical distinction between automated bug discovery and exploit weaponization. He warns that unfeasible regulatory standards and domestic policy instability risk handicapping American cyber defense while global competitors advance unrestricted.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 9.8% of the talking time here. How this is scored →
Alex as informed peer 3.3 Guest teaching 6.8 Guest disagreement 3.0 Alex pushing back 2.2
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 9:16 Refusing the host's debate resolution
Stamos flatly rejects Kantrowitz's assertion that Ranjan won the debate, asserting that neither host fully grasped the core issue regarding bug-finding standardisation.
Hardest push from Alex ▶ 11:30 Kantrowitz challenges the hype dismissalKantrowitz pushes back against dismissing Anthropic's actions as marketing, arguing that genuine frontier risk would look identical to restricted access rollouts.
Biggest teaching moment ▶ 4:47 Explaining the vulnerability discovery timelineStamos educates the hosts on how bug finding reached superhuman capability well before Mythos, comparing scaling GPU compute to multiplying elite human bug hunters.
Alex holds their own ▶ 16:03 Kantrowitz queries mathematical impossibility of jailbreak immunityKantrowitz demonstrates technical depth by pressing Stamos on whether government demands for unjailbreakable frontier models are technically feasible.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| The Government Takedown of Anthropic's Frontier Models | 2 | 6 | 1 | 0 | Alex Kantrowitz opens by asking for a factual breakdown of why the government forced Anthropic to pull down Fable and Mythos. Alex Stamos delivers a detailed chronological explanation of Amazon's security research, AWS classified hosting, and the Commerce Department's export control order. | |
| Debating AI Danger: Breakthrough Capability vs. Marketing Hype | 3 | 8 | 3 | 3 | The hosts frame a debate between whether Anthropic's models represent genuine step-change danger versus coordinated marketing hype. Stamos schools both hosts on cybersecurity realities, explaining that superhuman bug finding crossed the Rubicon a year earlier and using the F-150 versus McLaren metaphor. | |
| The Dangers of Restricting Bug Finding in American Models | 3 | 7 | 4 | 2 | When Kantrowitz playfully declares Ranjan the winner of the debate, Stamos rejects the verdict and reframes the core issue. He warns that crippling American AI models from finding software bugs creates an enormous defensive liability. | |
| Gating Access, Global Competition, and the Chinese AI Advance | 5 | 7 | 3 | 5 | Kantrowitz pushes back to defend his thesis, arguing that legitimate capability leaps would naturally necessitate gated access and safety guardrails. Stamos acknowledges gating mechanisms like KYC but highlights the strategic error of kneecapping US models while open-weight Chinese alternatives like GLM 5.2 rapidly advance. | |
| Security in Practice: Glasswing, Project Athena, and Existing Tools | 5 | 7 | 2 | 3 | Roy and Stamos discuss practical initiatives like Glasswing and Project Athena, noting that most enterprise codebases have simply never been audited. Kantrowitz raises an informed technical question regarding whether proving a model is unjailbreakable is even mathematically possible. | |
| Political Instability and the Strategic Fallout of AI Takedowns | 2 | 6 | 5 | 0 | Stamos delivers a passionate monologue on the strategic fallout, arguing that sudden regulatory takedowns inject political risk into US cloud providers and force enterprise CTOs toward Chinese open-weight fallbacks. Kantrowitz listens and promptly invites Stamos back for an extended episode. |
Statements from this episode (17)
Amazon hosts Anthropic's AI models across classified government cloud environments
“Amazon hosts Anthropics models in the, in what they call bedrock. This includes in all of the classified spaces. So if the NSA is using Anthropics models. It is running in AWS secret cloud or top secret cloud.”
Commerce Department export controls barred foreign Anthropic developers from their models
“The Commerce Department signs an order saying that the models are designated export controlled and that no foreigner, including the Foreign citizens who worked and actually built the models themselves can touch them.”
Anthropic pulled Fable and Mythos rather than seeking a restraining order
“Anthropic decides to actually follow through instead of immediately getting a temporary restraining order, and they, because they have no ability to make sure that no foreign citizen is, is seeing the models actually pulls them down. Both Mythos which was priv…”
Anthropic's warnings around dangerous AI models are coordinated PR stunts
“I think the reason I get so suspect on it is how perfectly coordinate coordinated a lot of the marketing and communications are. Around how they roll these out. Like, again, mythos is the most dangerous thing. There's someone eating a sandwich in a park that a…”
Anthropic's Mythos leads all known public cybersecurity benchmark evaluations
“So mythos is almost it is. The best bug finding model that we know about for sure. It is most likely the best bug finding model in the world. And I say most likely because we do not know what the Chinese secretly have in their labs. But for all the stuff for w…”
Frontier AI models matched elite human bug finders last year
“This Rubicon was crossed well last year with the Opus four series with the GPT five series. That is when the models became better or as good or better than the best human bug finders and not just as good as the best Bug finders.”
GPT-5 and Chinese AI model Kimi readily find bugs and write exploits
“You can just ask GPT-V to go find these bugs and it will go do it for you. You can just ask Kimi, which is an open source Chinese model. It will go write you these exploits.”
Restricting AI bug finders disproportionately harms cyber defenders over attackers
“All you've done is actually hurt the defenders. Who want to have full coverage, right? Who want to be able to find all the bugs. An attacker only needs one or two bugs to string together to actually pull off the attack. And so that's why, you know, so about a …”
Banning security training for American LLMs will increase software vulnerabilities
“If our American LLMs know nothing about security, they will create more security flaws. So we cannot create a standard where American LLMs are dumb about security. That would be a humongous own goal.”
OpenAI is rumored to release a Mythos-quality cybersecurity model soon
“Open AI is rumored to be releasing something that's mythos quality really soon.”
Anthropic's Glasswing access gating excluded critical infrastructure operators
“In the case of glass wing, I think it was honestly it. I thought it was not widespread enough. I know of like critical infrastructure companies that don't have access to mythos right now.”
China's open-weight GLM 5.2 rivals top Anthropic and OpenAI models
“It is within percentage points of the top closed models from anthropic and open AI in a bunch of things a bunch of evals. We don't know how good it is at bug finding yet, so there hasn't been any good testing here, but encoding a bunch of other intelligence ta…”
Anthropic's Glasswing participants found 10,000 security bugs but fixed only 1,000
“They have found like 10,000 bugs. Now, one of the problems with Glasswing is they found like 10,000 bugs, but they've only fixed a thousand.”
Claude Opus 4.8 will catch 70% of important cybersecurity bugs
“Use Opus four eight right now, and I guarantee you'll get 70% of the bugs you care about.”
NIST research proves making an AI jailbreak-proof is mathematically impossible
“No, there's actually a paper from NIST. It's like a kind of a Godel completeness theorem kind of paper that basically says it's impossible to make a model that is completely not jailbreakable.”
Broad AI bug-finding bans will force US cybersecurity to Chinese models
“What would be terrible for the entire American AI and cybersecurity industries is if the Trump administration defines a jailbreak as having any ability to find any vulnerability in code, because then all of us have to go to Chinese models.”
CTOs are deploying Chinese open-weight models as domestic regulatory backups
“And so now this week, CIs and CTOs are signing contracts to have open weight models on different hosts, on US hosts. They're not using Chinese hosts, but they're using Chinese models on as backups through LLM routers because political risk Is now a risk of usi…”