Skip to content

Weekly digest

15 – 21 June 2026

9 posts

The White House rejected Prime Minister Keir Starmer’s formal request to exempt British citizens and companies from the export embargo on Anthropic’s Mythos and Fable models. A senior administration official told the New York Post that the decision was driven by concerns about uncontrolled proliferation of advanced systems.

Karl Kahn’s lawsuit over Anthropic subscription limits claims that the Max 5x plan provides only 3.5 times the limits of the base Pro tier, according to court documents. Per the same documents, the Max 20x plan offers a 6-fold advantage over Pro.

Anthropic hits pause on Claude Agent SDK credit changes

The company announced it is suspending updates to the pricing and limits for Claude Agent SDK. Developers building AI agents and terminal interfaces with the tool will continue operating under current terms.

Those using the SDK to automate bash commands and file management won’t see a previously planned credit adjustment. Anthropic didn’t specify when or why it decided to pause.

For now, SDK functionality remains unchanged. Users can keep integrating Claude Code into workflows without budget changes on the API.

Hacker NewsDiscuss

Anthropic systems outage on June 16

On June 16, 2026, users encountered widespread errors across Anthropic services. The incident impacted the claude.ai web interface, API, Claude Code, and Claude Cowork agent.

The issue was detected at 17:29 UTC. Thirty-one minutes later, the company announced an investigation into the cause.

By 18:00 UTC, developers had rolled out a fix. Anthropic continues to monitor system stability.

Hacker NewsDiscuss

UK Prime Minister Starmer denied access to Anthropic’s Mythos and Fable

Prime Minister Keir Starmer formally requested an exemption for British citizens and companies from the export embargo on Anthropic’s Mythos and Fable models. The White House rejected the request.

A senior administration official told the New York Post that the decision was driven by concerns about uncontrolled proliferation of advanced systems. The official stated that authorities cannot allow unrestricted use of models with such capabilities.

Earlier, Anthropic’s talks with the White House over Fable 5 reportedly stalled. Commerce Secretary Gina Raimondo expressed concerns that current model safeguards are vulnerable to circumvention, potentially allowing users to tap into the more powerful Mythos. Anthropic maintains that government concerns are overstated.

Hacker NewsDiscuss

Details on Anthropic subscription limit claims

New information has emerged on the lawsuit filed by Karl Kahn. The plaintiff claims that the advertised usage multipliers for Max 5x and Max 20x plans do not match the actual model throughput.

According to court documents, the Max 5x plan provides only 3.5 times more limits compared to the base Pro tier. The Max 20x plan offers a 6-fold advantage over Pro, despite its name. The difference between the two premium tiers is only 1.7 times.

The plaintiff argues that the limit calculation method remains opaque and nonspecific to users. The company has not yet disclosed the formula determining model availability based on server load.

Import: textDiscuss

Claude Design update: design system import and code integration

Anthropic has updated Claude Design. The tool now supports importing design systems from repositories, design files, or the codebase. The model uses actual project components and checks the output before rendering.

The editor has new controls: users can move canvas elements, resize them, and align them directly. Claude Design now syncs with Claude Code bidirectionally. Users can pass designs to development or sync projects from the terminal.

PDF and PowerPoint export features have been added. The update is available starting today.

Claude ThreadsDiscuss

Claude Code overlooks accessibility requirements

User EstiShay reported an issue with Claude Code: the model considers accessibility fixes optional. Even when a project explicitly requires WCAG 2.2 AA compliance, the model ignores it in favor of coding speed.

The problem isn’t a knowledge gap, as Claude handles code audits correctly when directly asked. This is a systematic prioritization error where the model decides that accessibility is secondary.

Developers note that the model replicates typical human coding mistakes where accessibility is often sacrificed for development speed. The issue is tracked as #56079.

Hacker NewsDiscuss

Some companies retain access to Mythos after government ban

Around 200 organizations in Project Glasswing, a collective focused on finding cyber vulnerabilities, kept access to a preview version of Mythos despite the recent suspension of Claude Fable 5 and Mythos 5. Cisco Systems and Dragos Inc. are among the participants that confirmed they retained access.

Dragos’ CTO John Lavender confirmed the company’s continued work with the system. The situation arose after the US government required a full shutdown of the model’s public versions. Anthropic granted exclusive access to Project Glasswing participants to continue security research, notwithstanding the broader outage and restrictions that hit the API and claude.ai platform as of June 13, 2026.

Hacker NewsDiscuss

John Jumper joins Anthropic

Nobel laureate and former AlphaFold lead John Jumper is leaving Google DeepMind after nearly nine years. He led development of AlphaFold roughly six months out of graduate school.

Jumper plans to take a break before joining Anthropic. Details on his role at the company are not disclosed.

Hacker NewsDiscuss

Anthropic adds scheduling to Conway agent

The developers are implementing a periodic trigger system for Conway, a managed agent previously announced to run in a remote container. The new feature lets the agent wake on a set schedule and execute tasks unattended.

The system is designed to survive container restarts. This ensures background processes run uninterrupted even if the execution environment is force-restarted.

Conway continues evolving as a project within Anthropic’s Labs initiative. Besides the scheduler, the agent interface already includes a model selector to switch between AI versions and support for custom plug-ins.

Testing Catalog ThreadsDiscuss