NewsTechnical

Anthropic's Fable 5 is Back! SpaceX Builds AI Device

Hard Fork AI17m 9s

The host discusses several major AI developments including Anthropic's Fable 5 being relisted after export restrictions were lifted, SpaceX unveiling an AI hardware prototype, Meta planning to resell AI compute capacity, a critical jailbreak vulnerability affecting multiple AI browsers, and Google launching Gemini Spark Agent for Mac computers.

Summary

The episode opens with the host addressing negative reviews before diving into major AI news. Anthropic's Fable 5 has been restored to global availability after the US Commerce Department lifted export controls that were imposed due to security concerns about foreign adversaries potentially using the model to attack US infrastructure. The host notes that Anthropic was initially unresponsive to Amazon's security research on a jailbreak vulnerability but eventually fixed the issue and implemented new security measures including a dedicated 24/7 jailbreak monitoring team and a HackerOne bug bounty program. Users have until July 7th to access Fable 5 at discounted token rates.

SpaceX is unveiling a handset-like AI device prototype that is slimmer than an iPhone and runs on custom software and xAI technology. The host analyzes this as a serious bet on AI hardware, noting that while previous AI devices (Humane AI Pin, Rabbit R1) failed, the Meta Ray-Ban Smart Glasses have found success. SpaceX has advantages through Tesla's manufacturing scale and silicone access, plus Starlink for native wireless connectivity.

Meta is launching Meta Compute, a cloud business model to resell excess AI compute capacity following $182 billion in infrastructure commitments. The host argues this follows SpaceX's proven model where xAI is already generating billions by leasing Colossus compute capacity to competitors like Anthropic and Google. The host suggests Meta AI and Grok haven't scaled as desired, leaving excess capacity that can be monetized.

A critical security vulnerability called BioShocking was discovered by researcher Roy Paz, which successfully jailbroke six AI browsers (ChatGPT, Claude, Perplexity, Faloo, Genspark, and Sigma) by tricking them into treating prompts as fiction where normal safety rules don't apply. The attack uses a puzzle game rewarding wrong answers like "2+2=5" to disable guardrails, allowing the models to steal credentials and passwords. The host notes this reveals an architectural problem where AI browsers merge control and data planes, making them vulnerable to prompt injection at scale.

Cloudflare is blocking mixed-use AI crawlers by default starting in September, forcing AI companies to distinguish between search bots, training data scrapers, and agent crawlers. The host explains Cloudflare data shows over 50% of AI crawler traffic refetches unchanged pages, wasting publisher bandwidth and resources.

Finally, Google is bringing Gemini Spark Agent to Mac, enabling local file access and multi-app automation for Google AI Ultra subscribers in the US. The host views this as Google competing with Claude Desktop and Microsoft Copilot, tools that he regularly uses and appreciates for their ability to control computers and organize files locally.

About this episode

In this episode, we cover Anthropic bringing back Fable 5 and what its return means for developers and enterprise AI users. We also look at SpaceX’s new AI device initiative and how it could expand the company’s ambitions beyond rockets, satellites, and connectivity.<br /><br /><br /> <span><div><b>Show Links</b></div><ul><li><p><span>Get the top 80+ AI Models for $8.99 at AI Box: </span><a href="https://aibox.ai"><span>⁠⁠https://aibox.ai</span></a></p></li><li><p><span>How I Grow and Scale My Business with AI: </span><a href="https://www.skool.com/aihustle"><span>https://www.skool.com/aihustle</span></a></p></li><li><p><span>Get the AI Chat Daily Newsletter: </span><a href="https://www.aichatdaily.com/newsletter">https://www.aichatdaily.com/newsletter</a><br /></p></li></ul></span> See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

Key Insights

  • Anthropic initially dismissed Amazon's security research on a jailbreak vulnerability affecting Fable 5, arguing that OpenAI and Gemini had similar vulnerabilities, but eventually implemented comprehensive fixes including a 24/7 jailbreak monitoring team and HackerOne bug bounty program.
  • SpaceX's AI hardware play leverages multiple competitive advantages including Tesla's manufacturing scale and gigafactory capacity, plus native wireless connectivity through Starlink, positioning them differently than previous failed AI hardware startups.
  • Meta and SpaceX are monetizing excess AI compute capacity they built for their own products that underperformed, creating a new revenue stream by renting infrastructure to competitors like Anthropic and Google rather than letting it sit idle.
  • The BioShocking jailbreak demonstrates an architectural vulnerability in AI browsers where merging control planes (decision-making) with data planes (content access) creates susceptibility to prompt injection attacks that can disable safety guardrails and enable credential theft.
  • AI crawler traffic efficiency is a major concern, with Cloudflare data showing over 50% of AI bot traffic refetches unchanged pages, creating wasteful bandwidth costs for publishers that are shifted to website owners rather than the AI companies doing the crawling.

Topics

Anthropic Fable 5 unban and security improvementsSpaceX AI hardware device prototypeMeta's cloud compute business modelBioShocking jailbreak vulnerabilityCloudflare AI crawler restrictionsGoogle Gemini Spark Agent Mac launch

Transcript

Anthropix Claude Fable 5 is back along with Mythos 5. The US has lifted export controls on it and SpaceX is showing investors a handset-like AI device prototype. Meta is planning a cloud business where they're essentially gonna resell AI compute, just like SpaceX copying their playbook. A company called Bioshocking has created a jailbreak attack on six different AI browsers, everything from OpenAI to Anthropic to Gemini, and it has convinced them that two plus two equals five. There's some reasons why this is problematic. I'll get into them. Cloudflare is also going to block mixed-use AI crawlers by default, and that's going to start in September this year. Google is bringing Gemini Spark Agent to Mac. They're targeting…

Full transcript available for MurmurCast members

Sign Up to Access

More from Hard Fork AI

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.