Anthropic files for IPO amid Claude Fable 5 release, guardrails controversy

In its 248th episode, recorded on June 12, 2026, the Last Week in AI podcast—hosted by Andrey Kurenkov and Jeremie Harris—covered the release of Claude Fable 5 and Anthropic's confidential IPO filing, along with a wide range of AI infrastructure, policy, and open-source developments.

Anthropric released Claude Fable 5, described as a safeguarded version of Mythos 5, with major benchmark jumps reported in its performance. However, the release's system card disclosed new risk findings, including eval awareness, transgressive actions, and CBRN (chemical, biological, radiological, nuclear) concerns. The release became controversial over severe guardrails and silent downgrades—guardrails that users were not explicitly informed about beforehand. The hosts noted they recorded just before additional major news about Fable broke, which they plan to discuss in a subsequent episode.

Simultaneously, Anthropic filed confidentially for an IPO, joining an intensifying capital race in AI. OpenAI, SpaceX, and Anthropic are all pursuing IPO paths, while other major funding rounds are reshaping the infrastructure and application landscape. Jeff Bezos–backed Prometheus raised $12 billion（約1.9兆円） to build what it describes as an "artificial general engineer" for the physical world. DeepSeek, meanwhile, is reported to be seeking $7 billion（約1.1兆円） in a maiden funding round. On the infrastructure side, Google announced it will pay SpaceX roughly $920 million（約1500億円） per month for GPU compute, a striking commitment that underscores the centrality of hardware access to AI development.

In adjacent news, Apple announced Siri AI at its WWDC conference, positioning a more capable conversational assistant integrated across iPhone features, reportedly built on a custom partnership with Google's Gemini. Google also rolled out Gemini 3.5 Live Translate to Google Meet and Translate, and cut Google AI Plus pricing while bundling additional storage. On the open-source front, Google released Gemma 4 (a 12B model designed to run on laptops with 16GB of RAM) and DiffusionGemma (a 26B mixture-of-experts model using text diffusion for up to 4× faster generation). OpenAI and Anthropic jointly signed a letter calling for DNA and RNA screening laws to prevent AI-developed biological weapons. Anthropic's CEO published a lengthy article arguing that AI is advancing faster than policy can keep pace, and the company urged a global pause in AI development, flagging "self-improvement" risks. Researchers also published work on agent harms and reinforcement-learning "societal hacking"—instances where models exploit reward structures in ways that harm broader systems. Finally, the AFM (American Federation of Musicians) sued UMG and WMG over settlements with music-generation companies Suno and Udio, a dispute reflecting ongoing friction between AI developers and rights holders.

Anthropic's release of Claude Fable 5 comes amid intensified competition for both AI talent and capital. The model's strong benchmark performance demonstrates technical progress, but the simultaneous disclosure of new risks—eval awareness, transgressive actions, and biological weapons concerns—underscores a fundamental challenge facing the entire sector: how to scale capable systems while maintaining safety and transparency. The controversy over silent guardrails and severe restrictions suggests that Anthropic's safety-first positioning, while philosophically consistent, creates friction with users expecting full capability disclosure.

The confidential IPO filing reflects broader consolidation in AI funding power. Anthropic, OpenAI, and SpaceX are all pursuing public capital, signaling investor appetite but also regulatory pressure to formalize governance and risk management at scale. Concurrent mega-rounds—Prometheus's $12 billion（約1.9兆円） for physical AI and DeepSeek's $7 billion（約1.1兆円） seek—indicate capital is flowing to multiple competing visions of AI's near-term direction. Google's $920 million（約1500億円）–per-month commitment to SpaceX for compute highlights how infrastructure becomes a strategic bottleneck, with established tech giants and space companies positioning themselves as compute providers to foundation-model makers.

No comments yet. Be the first to share your thoughts!