Just as last week was ending, the US government forced Anthropic to pull its two newest models, Fable 5 and Mythos 5, citing national security concerns after Amazon researchers allegedly found a way to bypass Fable 5’s guardrails.
Cybersecurity researchers have since signed an open letter calling the move dangerous, and Anthropic itself noted the same jailbreaks exist in other models. So is this a genuine security concern, or just the latest chapter in a messy relationship between Anthropic and the Trump administration?
On this episode of TechCrunch’s Equity podcast, hosts Anthony Ha, Sean O’Kane, and Rebecca Bellan unpack what the ban means for developers building on Anthropic’s platform and for anyone watching the IPO, why it might accidentally be good for the company, and more of the week’s headlines.
Subscribe to Equity on YouTube, Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod.
Chapters:
00:00 Intro
01:28 The US government pulls Anthropic’s Fable 5 and Mythos 5
03:57 Cybersecurity researchers push back on the ban
06:33 Could the ban actually help Anthropic’s business?
12:26 The UK’s sweeping social media ban for users under 16
21:34 SpaceX acquires Cursor for $60B in stock
28:53 Jeff Bezos’s $12B bet on physical AI with Prometheus
32:04 Outro
@doubledragon9530
June 19, 2026 at 7:34 pm
Opus 4.8 was significantly better than 4.7, I mean qualitatively better. It still got confused on long analyses, however, but Fable, man that thing never let go of its focus, whatever you were exploring, it stayed focused right there with you. I was in the middle of two substantial projects just three days in when they pulled it, and I those two projects are now on hold because, as good as Opus 4.8 is, I’m certain it won’t do as good a job and will require significantly more work from me to bring those to completion. Also, if anything needs to be slowed down, it’s the creation and use of agentic tools because that is where the real security risk is right now, poisoned tools. Check out exploits of Openclaw and vcode to name just two attack surfaces. Adversaries aren’t going after jailbreaking text prompts anymore, they are going after backdooring tools. All of that needs to be locked down, not the models themselves.