← Back to all entries
2026-09-01 🧭 Daily News

Music Publishers Sue, Agent Domain Controls, and Code Limits Walk-Back

Music Publishers Sue, Agent Domain Controls, and Code Limits Walk-Back — visual for 2026-09-01

🧭 Sony Music and Warner Chappell Sue Anthropic, Alleging a "Brazen Campaign" of Copyright Piracy

Sony Music Publishing, Warner Chappell Music, and a coalition of major music publishers filed suit against Anthropic in the U.S. District Court for the Northern District of California on August 29, 2026. The complaint accuses Anthropic of conducting a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works" to build Claude's training datasets — characterising the conduct as large-scale piracy rather than fair use.

Where this fits in Anthropic's mounting copyright litigation

This is the third major music-industry copyright action targeting Anthropic in 2026 alone:

Why the "acquisition method" theory matters

The Bartz ruling shifted the legal terrain significantly: fair-use defences may cover what AI companies do with copyrighted text during training, but they do not protect how that text was obtained. If Anthropic sourced content through torrents or scraping tools that circumvent access controls, those acts are independently actionable under the Computer Fraud and Abuse Act and DMCA Section 1201 — regardless of the model's downstream use. Each new lawsuit applying this framework increases Anthropic's litigation exposure and signals a coordinated industry legal strategy.

Anthropic said it will defend itself against the claims. The company has not publicly disclosed the data sources used to train Claude, a gap that prior courts have noted when ruling on discovery motions.

copyright lawsuit Sony Music Warner Chappell training data legal

🧭 Claude Managed Agents: Domain-Level Web Filtering and Live Tool Swaps Without Agent Rebuilds

The September 2026 platform release notes add two complementary capabilities to Claude Managed Agents that significantly tighten control over agentic web access at runtime — without requiring a new agent version for either.

1. Domain-level filtering on web tools

You can now restrict which sites a Managed Agents session's web_search and web_fetch tools can reach. Set allowed_domains or blocked_domains on the tool's entry inside the agent_toolset_20260401 configs array:

# agent definition excerpt
{
  "agent_toolset_20260401": [
    {
      "type": "web_search",
      "config": {
        "allowed_domains": ["docs.anthropic.com", "github.com"],
        "user_location": {"country": "US", "region": "CA"}
      }
    },
    {
      "type": "web_fetch",
      "config": {
        "blocked_domains": ["pastebin.com", "hastebin.com"],
        "max_content_tokens": 8000
      }
    }
  ]
}

For compliance-sensitive deployments — financial services, healthcare, government — this lets you lock an agent to a curated set of approved sources and prevent data leakage to arbitrary external endpoints, all declared at the tool-config level rather than enforced by prompt instructions (which can be bypassed).

2. Mid-session tool updates

You can now call the session-update endpoint to patch a running session's agent.tools and agent.mcp_servers — including their permission policies and per-tool web settings — without ending the session or creating a new agent version:

# Python SDK — update domain filters mid-session
client.beta.managed_agents.sessions.update(
    session_id=session_id,
    agent={
        "tools": [
            {
                "type": "web_search",
                "config": {"allowed_domains": ["internal.company.com"]}
            }
        ]
    }
)
# Updated allowed_domains apply immediately to the rest of this session
Important scoping rule

Mid-session updates are session-local — they do not write back to the underlying agent definition. The next session started from the same agent will use the original tool config. This is intentional: it lets you apply one-off overrides for a specific run (e.g., temporarily widening access for a debugging session) without permanently altering the agent's configured behaviour.

Managed Agents domain filtering web_search web_fetch mid-session enterprise compliance

🧭 Anthropic Corrects Its Claude Code Limits Post — 17% Weekly Cut Arrives September 14

The +50% weekly usage boost that Claude Code Pro, Max, Team, and legacy seat-based Enterprise users received on May 13, 2026, expired at 11:59 PM PT on August 31 — its third extension having run its course. What users didn't expect was Anthropic's follow-up correction.

The retraction

Anthropic initially posted that weekly limits would simply return to their pre-boost baseline once the extension ended, implying a flat reset. The post was deleted and republished with a correction: limits will not merely revert — they will drop 17% below the pre-boost baseline, effective September 14, 2026. The two-week gap between September 1 and September 14 gives users time to adjust pipelines and evaluate usage patterns before the cut lands.

What the numbers mean in practice

The +50% boost temporarily pushed weekly limits to 1.5× their prior baseline. On September 1 they return to 1.0×. On September 14 they fall to approximately 0.83× of that same baseline — meaning the net drop from the boosted peak is about 45%.

What to do before September 14

Run /usage in Claude Code to see your current weekly consumption. If your workflows routinely hit 70–90% of the weekly limit, schedule compute-heavy sessions for before September 14, and consider whether API-direct access with explicit token budgets via the Claude API pricing page is a better fit for predictable high-volume work. For Team and Enterprise admins, this is a good moment to audit seat usage and confirm your per-seat allocations are sized correctly for the lower baseline.

Claude Code rate limits usage Pro Max Team Enterprise
Source trust ratings ⭐⭐⭐ Official Anthropic  ·  ⭐⭐ Established press  ·  Community / research