The AI Show: Anthropic's Fable 5 Pullback, OpenAI's IPO, and Apple's Siri Overhaul

This week on The Artificial Intelligence Show, Paul Ritzer and Mike Kaput dive into a whirlwind of AI news, from government intervention in model releases to major company IPO filings and long-awaited product updates. The conversation kicks off with the dramatic withdrawal of Anthropic's Claude Fable 5 and Mythos 5 models due to national security concerns, followed by an in-depth look at OpenAI's confidential IPO filing and Apple's significant AI-powered Siri upgrade. The episode also touches on the evolving cost of AI, the potential shift away from offshoring, the rise of "loops" in AI agent interaction, and the future trajectory from AGI to ASI.

Anthropic's Fable 5 and Government Intervention

The most immediate and dramatic news of the week revolves around Anthropic's Fable 5 and Mythos 5 models. Fable 5, Anthropic's new flagship model, was touted as a major leap in capabilities, with early reports highlighting its prowess in coding, vision, and long-context tasks. The payments company Stripe, for instance, claimed Fable 5 compressed months of engineering work into days, completing a 50 million line code migration in a single day.

However, just days after Fable 5's release, the U.S. government issued national security export controls, barring Anthropic from distributing these models to foreign nationals. This action forced Anthropic to disable the models for all users, as selective filtering based on citizenship proved unfeasible. The government's intervention reportedly stemmed from researchers at Amazon discovering a method to "jailbreak" Fable 5, potentially enabling cyberattacks. Amazon CEO Andy Jassy is said to have personally raised these findings with senior administration officials.

Anthropic has disputed the severity of the vulnerabilities, stating they were minor and replicable with other publicly available models, and that no universal jailbreak has been found. Despite this, Anthropic's technical staff traveled to Washington to meet with White House officials, and as of this recording, both models remain unavailable.

This situation has significant implications, as Anthropic warns that such stringent standards could halt all new model deployments for frontier model providers. The incident also highlights the growing tension between rapid AI advancement and the slower pace of policy and regulation.

Dario's Essay and Policy Proposals

Prior to the government's intervention, Dario Amodei, CEO of Anthropic, published an essay titled "Policy on the AI Exponential." In it, he addressed the rapid pace of AI development, noting that models have evolved from barely writing coherent code to writing most of the code at major AI companies in just four years. He emphasized the mismatch between AI's exponential growth and policy's slow progression, warning that "in the several years that it can take Congress to act, AI can go from an amusing toy to the full country of geniuses."

Amodei outlined five core areas for regulation and public safety: regulation and public safety, macroeconomics and tax policy, scientific innovation, balance of power between state and society, and geopolitics. Within the regulation and public safety section, he proposed that frontier AI models, like airplanes, should undergo technical testing and auditing, with releases blocked or reversed if they don't meet high safety standards. He specifically suggested that the government should have the power to block or deter deployment if third-party assessments deem a model to present unacceptable risks, with safeguards against political favoritism.

Anthropic also released two policy proposals: an advanced AI framework for governing increasingly capable systems and an economic policy framework addressing potential job displacement. These frameworks explore scenarios for unemployment ranging from 5% to unprecedented levels, and discuss the economic transition ahead, particularly in the context of recursive self-improvement.

OpenAI Files for IPO

In other major news, OpenAI has officially kicked off its initial public offering (IPO) process by filing the necessary paperwork confidentially. The company is reportedly targeting a public listing within the next year, with a valuation around $850 billion. OpenAI plans to transform ChatGPT into a "super app" by integrating codecs, AI agents, and various other features to drive revenue.

Sam Altman, OpenAI's CEO, has suggested that the pace of recursive self-improvement in AI could influence the IPO timing, stating that a faster "RSI takeoff" might make delaying the IPO advantageous. This indicates a strategic approach to timing the market, potentially waiting for their next model release to be on par with or surpass competitors like Anthropic.

OpenAI also published a post titled "Built to Benefit Everyone: Our Plan," outlining its third phase of development. This phase focuses on reshaping the economy around AI, with the goal of building an automated AI researcher, accelerating scientific progress and economic growth, and providing everyone with personal AGI. The messaging emphasizes that AI should work for people, augmenting their capabilities rather than entirely automating tasks, and that human judgment and values remain crucial.

Apple's Siri AI Is Finally Here

After years of speculation and anticipation, Apple has unveiled a significantly overhauled Siri, now powered by AI. Dubbed "Siri AI," the new virtual assistant is designed to be more conversational, knowledgeable, and capable. It can understand on-screen context, search across personal data like messages, emails, and photos, pull in real-time web information, and take actions across apps and devices. Siri AI will maintain a conversation history in a dedicated app, synced privately via iCloud, and will be available across Apple's ecosystem, including iPhone, iPad, Mac, Apple Watch, and Vision Pro.

While a beta version is expected later this year for English-speaking users, Apple has not provided a firm release date for the full version. Analyst Gene Munster estimates a full release could be as late as spring 2027, but notes that Apple has a significant advantage as competitors currently lack compelling personalized AI. Beyond Siri, Apple also announced new parental controls, screen time features, and performance enhancements for its apps.

Is the Era of Affordable AI Over?

A recent analysis from SemiAnalysis suggests that the era of cheap AI might be coming to an end. The firm tested subscription plans from Anthropic and OpenAI, finding that the generous token allowances on plans like Claude's $200/month offering delivered up to $8,000 worth of tokens at API pricing, and a $200 ChatGPT plan yielded about $14,000. This indicates that AI labs have been heavily subsidizing their power users.

As companies transition to pay-as-you-go models, costs are beginning to rise. Meta, for example, is reportedly curbing employee AI usage due to escalating internal AI costs, which have reached billions. OpenAI is reportedly considering drastic price cuts to preempt a user war with Anthropic, but the fundamental constraint remains the immense compute power required for these models. This situation creates significant uncertainty around AI pricing, with potential for substantial increases in per-seat costs. Google, with its profitable and financially secure position, is seen as having the most flexibility in navigating these pricing challenges.

Opendoor Ends Offshoring for AI-Native Workers

Opendoor, a major real estate company, is winding down its India-based operations and bringing that work back to the United States. CEO Cass Neaten framed this move as bringing operational work closer to their American customers. The company has unified its systems and hired AI-native, customer-facing teams in the U.S., making in-person work closer to customers more efficient. This shift reflects a broader trend where the increasing power of AI agents reduces the need for offshoring manual workflows, allowing companies to consolidate operations and leverage AI tools to enhance productivity and impact.

From Prompts to Loops: The Next Frontier of AI Interaction

A new paradigm is emerging in AI interaction, particularly for coding agents: moving from direct prompting to designing "loops" that prompt the agents. This involves creating small systems that automate the prompting process, allowing agents to iterate on tasks, evaluate their own output, and continue prompting until a goal is met. This approach, championed by figures like Peter Steinberger and Boris Churnney, signifies a shift towards agents performing more of the work autonomously, with human roles evolving to designing and managing these automated loops. While currently at the frontier of AI usage, this trend could fundamentally change how average users interact with AI assistants in the future.

From AGI to ASI: The Path to Superintelligence

A new paper from Google DeepMind, titled "From AGI to ASI," explores the potential trajectories from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI). The paper, co-authored by DeepMind co-founder Shane Legg, outlines four potential pathways: scaling current AI, new AI paradigm shifts, recursive self-improvement, and ASI emerging from large groups of AI agents. It also discusses potential frictions that could slow down these advancements. The paper highlights the uncertainty surrounding AI's future progress and emphasizes the need for a massive, interdisciplinary effort to prepare for a future where AI capabilities could significantly surpass human organizations. Notably, the paper includes "summary instructions" for AI assistants, a novel approach to guiding AI in understanding and summarizing research.

Europe 2031: A Warning on AI Stagnation

A report titled "Europe 2031: What Getting AI Wrong Means for Us" paints a stark picture of Europe sliding into economic and geopolitical irrelevance by 2031 due to missteps in AI adoption. The report, authored by AI researchers, think tankers, and investors, identifies three key mistakes: underestimating AI's speed of development, misjudging its transformative potential, and overestimating Europe's ability to catch up. The scenario suggests that by 2031, the U.S. could possess over 12 times Europe's compute capacity, leading to significant economic consequences for the continent, including loss of competitive advantage, reduced tax bases, and rising unemployment. The core argument is that Europe needs to course-correct rapidly to avoid obsolescence.

AI Use Case Spotlight: Activating Research with AI

Smarter X's Director of Research, Taylor Ray, spearheaded the creation of the 2026 State of AI for Business report, which involved analyzing over 100,000 data points. The team then faced the challenge of activating this extensive research for their internal teams. To address this, they developed a Claude project designed to generate custom internal activation briefs for each of their eight business units. By loading the project with context about each unit's goals and priorities, the AI could map the report's findings directly to specific team needs, providing actionable insights and "quick wins." This approach highlights the power of AI in not only generating content but also in facilitating its practical application and activation across an organization.

AI Product and Funding Updates

This week's product and funding updates include:

Key Takeaways