Anthropic Disabled Fable 5 And Mythos 5 After A U.S. Export-Control Order. Here's What Happened Claude Fable 5 launched June 10 as the highest-scoring AI model ever made public and the U.S. Commerce Department shut it down within days after a coordinated multi-agent jailbreak extracted a drug synthesis route and security exploit guidance. As of June 16, Anthropic is in active negotiations over a tiered access framework. For CRE teams running workflows on Claude: Opus 4.8 is the stable fallback and should be tested now. The bigger lesson — 16% of companies have no contingency plan if an AI vendor goes offline. This week proved that risk is real.
AppFolio Connects Realm-X to Anthropic's Claude — A New Standard for Real Estate Performance Management Announced June 9, AppFolio built an agent-to-agent connector linking its Realm-X AI suite directly to Claude. Property managers can now trigger and execute complex operational work through Claude while AppFolio's compliance guardrails stay intact.
Gemini 3.5 Pro Nears June GA With 2M Token Context Window and Deep Think Reasoning Google's Gemini 3.5 Pro, announced at I/O in May, is in limited preview with general availability targeted for late June. For CRE: a 2M token context window means you can feed an entire portfolio's worth of leases, T12s, and lender packages into a single prompt. The analysis use cases are significant.
Markdown files store what Claude should know (your business context, current projects, key people, open decisions, property details). Think of them as your memory layer. CLAUDE.md is the main one, but you can build out a whole memory folder: people.md, projects.md, cap-table.md, whatever's relevant to your work. These files change constantly as your business evolves.
Skill files define how Claude should execute. These are step-by-step instructions for a repeatable workflow that gets triggered by your intent. "Analyze this T12." "Run the business review." "Pull newsletter content." The skill file tells Claude exactly what to do, in what order, and what to output. These stay stable once they're built.
The practical difference:
You update your markdown files every week. You update skill files maybe once a quarter.
Markdown files answer "what's true right now." Skill files answer "how do we do this."
Markdown files are passive as Claude reads them for context. Skill files are active and Claude runs them like a playbook.
Where it gets powerful: a well-built skill pulls from your markdown memory mid-run. So your T12 Analyzer knows your underwriting criteria. Your deal analyzer knows your return hurdles. Your business coach knows your current revenue targets. You're not re-explaining context every time as it's already there.
Every month, we host a live call walking through what changed in Claude and what to do about it in your CRE work.
This month had a lot. Five updates dropped in 30 days.
1. Claude Fable 5 (June 9) — New model tier above Opus, available to all paid users (blocked for now). Ranked #1 on Hebbia's senior-level financial analysis benchmark. The advantage grows with task complexity. Reach for it on full diligence packages, multi-asset portfolio underwriting, or anything where you're synthesizing multiple documents in one session. One caveat: Fable 5 stores conversations for 30 days, including for ZDR accounts. Keep confidential deal data on Opus 4.8 until your firm reviews the retention terms.
—> Yes this has been blocked but we have a feeling it’ll be making a comeback.
2. Claude Opus 4.8 (May 28) — Updated flagship at the same price as Opus 4.7. Three additions: Adaptive Thinking (Claude scales reasoning depth to the task automatically), Fast Mode at 2.5× speed and ⅓ the cost, and a 1M token context window, which is roughly 1,500 pages in a single session. Default model for most CRE work.
3. Usage windows extended — Rolling windows doubled from 5 hours to 10 hours. Peak-hour restrictions removed for Pro and Max. API rate limits scaled up. No action required, applies automatically. Biggest payoff for long diligence sessions, portfolio reports, and automation pipelines that were hitting limits mid-run.
4. Dynamic Workflows, Dreaming & Outcomes — Three Claude Code features. Dynamic Workflows runs hundreds of parallel subagents in one session — T-12 abstraction across 50 properties at once instead of one at a time. Dreaming reviews past sessions, spots recurring errors, and updates memory so the agent improves between runs. Outcomes lets you define a quality rubric; a separate Claude instance grades the output and sends it back for revision until it passes before you see it.
5. Scheduled Tasks (June 9) — Claude Code tasks now run on Anthropic's cloud infrastructure on a cron schedule. No local machine required meaning you can close your computer and this will still work (finally). Set it once: nightly portfolio digest, weekly lease expiration flag, monthly investor report. Claude initiates the work. You review the output.
Recording is in the community if you missed it.
Don’t miss our next call in the Studio, where one of our members Fly Jamerson is going to walk us through how she gets Claude to sound like her.
Building Your AI Voice Profile
Date: Friday, June 19th
Time: 1:30PM-2:15PM EST
Join before the call with your 7 day free trial.