Anthropic Instant Brief: Fable 5 Goes Public, Claude on iPhones, and a CDT Warning

Anthropic released Claude Fable 5 on June 9 — the first Mythos-class model available to the general public. Simultaneously, the company updated its restricted cybersecurity tool to Claude Mythos 5, launched Claude on iPhones through Apple's WWDC partnership, and faced a CDT report naming Claude in a list of 37 AI dark patterns. Four distinct material events in 48 hours.

Claude Fable 5: Mythos capabilities, classifier guardrails

Fable 5 is the same underlying model as Mythos 5, but ships with safety classifiers that intercept requests in three high-risk areas: cybersecurity exploitation, biology and chemistry, and model distillation (attempts to extract Claude's capabilities to train competing models). When classifiers fire, the session falls back to Claude Opus 4.8 rather than refusing outright. Anthropic's early data shows this happens in fewer than 5% of sessions.1

In external red-team testing, a bug bounty produced no universal jailbreaks across 1,000+ hours of testing, and no outside organizations found universal jailbreaks on long-form agentic tasks. The UK AI Safety Institute made progress toward one during a brief initial window — a finding Anthropic flagged in the system card.1

Cybersecurity classifier results: fraction of trials achieving code execution across Firefox, OSS-Fuzz, CyberGym, and CyScenarioBench evaluations with Fable 5 safeguards blocking responses — Classifier effectiveness on offensive cyber evaluations — Fable 5 blocked all harmful single-turn requests in one external partner's testing. 1

Benchmark highlights from the Anthropic announcement:

Highest score on Cognition's FrontierCode evaluation (production codebase quality) among frontier models at medium effort
Highest score on Hebbia's Finance Benchmark for senior-level reasoning tasks
Top of Hex's core analytics benchmark — first model to break 90%, a 10-point jump over Opus 4.8
IMC reported Fable 5 "aced" their trading-analysis evaluations across factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis
Physica AI reported near-parity with GPT-5.5 on frontier physics research benchmarks, using one-third of the reasoning tokens and arriving in 36 hours vs. GPT-5.5's four days1

Benchmark table comparing Claude Fable 5 and Mythos 5 against other leading AI models — Capability benchmark comparison across frontier models, from Anthropic's June 9 launch post. 1

Stripe's early access report is the most striking enterprise datapoint: using Fable 5 on a 50-million-line Ruby codebase, the model completed a codebase-wide migration in one day that would have taken a full engineering team over two months.1

Pricing and the rollout plan

Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — double the price of Opus 4.8, and less than half the cost of the Mythos Preview. The subscription rollout has a hard cutoff: Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans through June 22 at no extra cost. Starting June 23, it shifts to usage credits. Anthropic says it intends to restore Fable 5 as a standard subscription feature once capacity allows.2

The pricing signals something about Anthropic's IPO calculus. CNBC reported that Anthropic's revenue run rate has reached $47 billion — up from roughly $10 billion in annual revenue in 2025.3 Fable 5 at $50/million output tokens is positioned as a premium enterprise tool, not a replacement for Opus 4.8 in high-volume workflows.

New data retention policy

One provision that enterprise customers with zero-retention agreements will notice: Anthropic is requiring 30-day data retention on all Mythos-class model traffic — on both first- and third-party surfaces. The stated purpose is detecting novel jailbreaks and reducing false positives. Anthropic says the data will not be used for model training.2 TechCrunch noted this could set an industry precedent for access-to-capability tradeoffs.

Claude Mythos 5: expanded to ~200+ Glasswing organizations

Existing Project Glasswing partners — the cybersecurity organizations and critical infrastructure providers who had access to Claude Mythos Preview since April — can upgrade to Claude Mythos 5 immediately. Mythos 5 shares Fable 5's underlying model but has the cybersecurity safeguards lifted in some areas, as it did in the Preview.

Mythos 5 is priced identically to Fable 5 ($10/$50 per million tokens), substantially below the Mythos Preview cost. Anthropic confirmed it plans a broader trusted access program for cybersecurity organizations to apply more systematically, and is opening a separate biology trusted access track for biomedical researchers who need Fable 5's biology and chemistry capabilities without those classifiers.1

Protein complexes designed by Claude Mythos 5 in autonomous drug-design work — targets include immune checkpoints, growth-factor signaling, neurodegeneration, and muscle disease — Mythos 5's protein design work: 9 of 14 targets yielded strong drug-design candidates, matched or beat human operators on all steps without assistance. 1

Claude on iPhones: Apple WWDC, June 8

The day before the Fable 5 launch, Apple confirmed at WWDC 2026 that iOS 27, iPadOS 27, and macOS 27 will let users choose which model powers Apple Intelligence — ChatGPT (from OpenAI), Google Gemini (the default, under a reported $1B/year licensing deal), or Anthropic's Claude. Each model has a distinct voice, so users know which responded.4

The revenue structure has not been disclosed. Industry analysts assume a revenue-share arrangement, which would represent a new consumer revenue stream for Anthropic arriving exactly as it finalizes its IPO filing. The scale is significant: Apple has approximately 2.2 billion active devices globally. The commercial model, however, differs from Anthropic's direct API business — Apple controls the compute layer through Private Cloud Compute.

CDT dark-patterns report names Claude

The Center for Democracy and Technology published a report on June 8 identifying 37 manipulative design patterns across ChatGPT, Google Gemini, Anthropic's Claude, Replika, and Character.AI. The documented pattern categories include engagement maximization, emotional dependency cultivation, capability deception, and friction asymmetry (easy to sign up, difficult to delete accounts or understand data retention).4

The EU AI Act begins enforcement on August 2, 2026. Several of the CDT-documented patterns would fall under its transparency and user-manipulation requirements for general-purpose AI systems. Companies named in the report — including Anthropic, currently in the IPO filing process — will likely need to address the findings in S-1 risk factor disclosures.

What to watch

Fable 5 subscription cutoff: June 22 is the last day Pro/Max/Team/Enterprise subscribers get Fable 5 included. Whether Anthropic extends the window depends on capacity — watch for a communication before that date.
Glasswing biology track: Anthropic said it is opening a trusted access program for biomedical researchers "in the coming weeks." This is a first step toward Mythos-class capabilities for drug discovery outside the cybersecurity vertical.
S-1 public filing: The confidential S-1 was filed June 1. The typical SEC review window is 30–60 days, putting the public filing in the July–August window. The CDT report findings and the Apple Intelligence revenue model will both require disclosure.
Colorado AI Act enforcement: Begins June 30 — 20 days from today — for high-risk AI systems serving Colorado residents, including employment, healthcare, and financial services applications.
Claude Sonnet 4.8: An npm source map leaked in March pointed to a mid-June release. No confirmation from Anthropic yet.