150+ Security Leaders Sign Open Letter to Reverse the Fable 5 Ban
The backlash to the export controls just got a heavyweight roster. More than 150 cybersecurity practitioners and executives — including the CEOs of Adobe, Sophos, and Zoom, former Facebook security chief Alex Stamos, Bugcrowd’s Casey Ellis, cryptographer Jon Callas, and Luta Security’s Katie Moussouris — signed an open letter urging the Trump administration to rescind the directive that pulled Fable 5 and Mythos 5 worldwide.
Their argument cuts at the ban’s premise: the restriction “has taken the best models away from defenders” who use them to find vulnerabilities and harden software, while doing nothing to slow attackers. “To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous,” the letter reads. Moussouris, who reviewed the Amazon research, says what was described “wasn’t a real jailbreak” at all.
“Fix This Code”: The Three Words Behind the Ban — and Where Talks Stand
Fortune traced the origin to a private Amazon research paper that claimed a way to jailbreak Fable, reportedly surfaced to the White House as a national-security flag. The administration’s account is blunt: it asked Dario Amodei to fix the jailbreak or de-deploy the model, and he declined. Security researchers who’ve since read the paper say the underlying claim doesn’t hold up.
No reversal yet. Anthropic’s technical staff are holding meetings with the Commerce Department and the National Cyber Director’s office to unwind the control. Fable 5 and Mythos 5 stay suspended worldwide — even for Anthropic’s own non-American staff — while Opus 4.8 and the shipping lineup keep running. Production workloads are untouched; it’s the frontier tier that’s frozen.
Claude Logs 10th Disruption in 12 Days as Infrastructure Strains
Monday brought another one: elevated error rates on Claude Opus 4.8 and Haiku 4.5 that persisted past an afternoon fix attempt, marking the tenth significant disruption since June 5. Anthropic acknowledged its infrastructure has been “stretched by unprecedented demand” — a candid admission, but cold comfort if your app went sideways mid-afternoon.
The recurring gap: still no published root-cause analyses or engineering post-mortems for any of June’s incidents. For teams routing production traffic through a single model vendor, the cadence is a real continuity signal. If you haven’t wired up retries, timeouts, and a fallback model, this is the week to do it.
Managed Agents Add Dreaming, Multiagent Orchestration, and Webhooks
Anthropic keeps building out the agent platform. The latest round of Managed Agents updates adds “dreaming” (background reflection between tasks), multiagent orchestration, outcomes, and webhooks — plus scheduled deployments, vault-stored environment-variable credentials, and richer session-thread webhook events. It pairs with this week’s self-hosted sandbox and private-MCP beta for regulated teams.
Taken together, these are the controls that move agents from “cool demo” to “runs unattended at 3am.” Scheduled deployments and webhooks mean an agent can wake on a cron or an event, do work, and hand off — without a human babysitting the loop. Credentials in a vault, not a prompt, is the kind of detail enterprise security reviews actually check.
Reminder: Sonnet 4 and Opus 4 Are Now Retired
Monday’s deprecation deadline has passed. Calls to claude-sonnet-4-0 and claude-opus-4-0 now return errors with no silent fallback. If something in your stack has been intermittently failing this week, separate it from the outage noise — a dead model ID fails every time, not just during an incident.
The fix is still one line: repoint to claude-sonnet-4-6 and claude-opus-4-8, keep your key, re-run evals, redeploy. Grep your configs, CI jobs, and any unattended scripts before the next on-call rotation inherits the mystery.
Claude for Legal: 20+ Connectors and 12 Practice-Area Plugins
Anthropic’s legal push keeps compounding. Its Claude for Legal suite now spans 20+ MCP connectors and 12 practice-area plugins covering research, contracts, discovery, matter management, and legal aid. The connector list reads like the legal-tech stack itself: Harvey, Solve Intelligence, Definely, DocuSign, Ironclad, Thomson Reuters, Trellis, and the Free Law Project among them.
The strategy mirrors the broader MCP play — meet professionals inside the tools they already live in rather than asking them to come to a chat box. Legal teams became the most engaged Cowork cohort of any knowledge-work function after February’s first plugin, and this build-out is Anthropic doubling down on a vertical that’s clearly biting.
Snowflake Ties a $200M Knot to Put Claude in Front of 12,000 Enterprises
The distribution engine keeps running. Anthropic’s multiyear, $200M partnership with Snowflake embeds Claude models directly inside the data cloud, exposing them to more than 12,000 global enterprise customers without a separate procurement cycle. It sits alongside the recent DXC and TCS deals in the expanding Claude Partner Network.
The throughline is unmistakable: Anthropic isn’t only selling model access, it’s wiring Claude into the data and services platforms that big buyers already run on. Meet the workload where the data already lives, and adoption stops being a sales motion and starts being a default.
The Outages Are the Story Anthropic Should Worry About Most
The Fable 5 fight is louder, but ten disruptions in twelve days is the headline with the longer tail. Frontier capability is what wins the demo; reliability is what keeps the contract. Every enterprise buyer doing diligence right now is screenshotting the status page, and “unprecedented demand” reads very differently to a CISO than it does to a growth investor — one hears scale, the other hears single point of failure.
The absence of public post-mortems compounds it. Mature infrastructure providers earn trust precisely when things break, by showing their work afterward. Until Anthropic publishes root-cause analyses with the same confidence it ships models, the gap between “most capable” and “most dependable” stays open — and that gap is exactly where multi-model strategies get justified in budget meetings.
When the Defenders Side With the Vendor, the Ban Has a Problem
The open letter changes the shape of the Fable 5 dispute. This isn’t Anthropic lobbying for its own product — it’s 150+ independent security leaders, including rival CEOs, arguing the government’s own safety rationale is backwards. When the people whose job is to red-team models say the “jailbreak” wasn’t real and the ban hurts defense, the policy loses its expert cover.
That doesn’t mean the control gets reversed tomorrow — national-security decisions rarely move on technical merit alone. But it raises the cost of holding the line, and it reframes the precedent: if a contested research claim can pull a shipped US model, every lab now prices regulatory risk into launch day. The resolution here will outlast the news cycle, because it sets the rules for whoever ships next.