Sony Music Publishing and Warner Chappell Sue Anthropic — and Name the Founders
Filed Friday, August 28 in the US District Court for the Northern District of California: Sony Music Publishing and Warner Chappell Music accuse Anthropic of a “brazen campaign of illegally torrenting, scraping and downloading copyrighted works on a massive scale” to train Claude. The publishers call it one of the largest ongoing thefts of intellectual property in history. Anthropic rejects the allegations and says it will defend itself in court.
Two details make this different from the music-industry suits that came before it. First, the complaint names Dario Amodei and Benjamin Mann as individual defendants, not just the company — an escalation that is rare and deliberate. Second, the publishers are seeking a jury trial with statutory damages of up to $150,000 per infringed composition across “tens of thousands” of songs, which is how you get to a multi-billion-dollar number without saying one.
The underlying theory is the one worth watching: the complaint alleges Claude can reproduce copyrighted lyrics in its outputs, which would let it substitute for licensed lyric services and undercut that market directly. That is a harder argument to wave off than training-data provenance alone. Anthropic already operates under court-approved guardrails from the 2023 Universal Music Group, Concord and ABKCO suit requiring it to prevent unauthorized lyric reproduction — so the existence of those guardrails is now evidence in both directions.
Claude for Teachers Goes Free for Entire School Districts
Announced August 28: Claude for Teachers is now available to US K-12 schools and districts as a free Enterprise offering, extending the individual-educator program Anthropic launched in July. Organizations that sign up by June 30, 2027 get a full year of free access.
The mechanics are standard enterprise onboarding with a K-12 wrapper. An administrator gets verified, accepts the K-12 terms and a student data privacy agreement, then connects an email domain and single sign-on. From there they manage staff accounts through role-based access control, set policies, and see adoption across schools. Educators get the teaching skills, the state standards and curricula underneath them, and connections to existing K-12 tools.
The part that actually unblocks procurement is the paperwork: Anthropic is providing district-level K-12 terms and a Data Processing Agreement aligned to FERPA, so one set of commitments covers the whole organization instead of each teacher agreeing to consumer terms individually. Free-through-2027 is a land grab, and everyone involved knows it — but the compliance scaffolding is the thing districts could not build themselves.
Anthropic Moves to Let Enterprises Hold Their Own Retention Data
Anthropic plans to let business customers keep greater control of their data on its most capable models — a reversal of the retention policy it announced in June, which required retaining all customer data for the Mythos and Fable models and future frontier systems as a cyberattack mitigation.
The revised approach keeps the 30-day retention requirement but adds the option to satisfy it on the customer’s own cloud infrastructure rather than Anthropic’s. A new safety system enabling this is expected to roll out later this year, developed in coordination with more than 100 customers in highly regulated industries.
What makes this notable is that Anthropic said the quiet part first. Its own reporting acknowledged the original policy would “be unpopular with customers who have come to expect zero retention, and pose real risks to our business success.” That is a company writing down the cost of its own safety decision before the market forced it to. OpenAI is testing a parallel approach it calls private safety processing, with wider release planned for September — so the entire frontier is converging on retain, but not here.
Enterprise-Managed MCP Authorization Hits GA, Adds Datadog, Notion and Slack
Enterprise-managed authorization for the Model Context Protocol connector framework is now generally available. The update brings Datadog, Notion and Slack into the supported roster, joining Asana, Atlassian, Canva, Figma, Granola, Linear and Supabase. Exa, Miro and Zoom are slated to follow.
The problem this solves is unglamorous and enormous. Until now, connecting Claude to a company’s internal tooling generally meant individual users completing individual OAuth flows, which produces a permissions surface no security team can audit and no admin can revoke centrally. Enterprise-managed auth moves that decision to the organization: the admin authorizes the connector, and user access follows existing identity and group membership.
Read this next to the Compliance API reaching Cowork transcripts and the domain allowlists on Managed Agents, and a pattern emerges. Anthropic spent this month shipping almost nothing that makes Claude smarter and almost everything that makes Claude approvable. That is what a company does when the bottleneck stops being capability and starts being procurement.
Managed Agents Get Session Budgets, Geo Pinning and Domain Allowlists
The Claude Developer Platform added a cluster of governance controls for Managed Agents: session budgets, advisor models, inference geo pinning for data residency, and GitHub-hosted skills. Separately, Admin API user-management endpoints for Claude Enterprise organizations — members, invites, groups and custom roles — reached general availability.
The most immediately useful of these is the smallest. You can now restrict which sites an agent’s web_search and web_fetch tools are allowed to reach by setting allowed_domains or blocked_domains on the tool’s entry in the agent_toolset_20260401 configs array. For any agent processing untrusted input, a network boundary is the single highest-leverage control available — it turns prompt injection from “the agent can be told to exfiltrate” into “the agent can be told to exfiltrate to a host it cannot reach.”
Inference geo pinning is the enterprise-sales unlock. Data residency is the objection that ends European and public-sector deals before pricing comes up, and it is not a thing you can negotiate around. Session budgets are the answer to the other recurring fear about long-running agents: not that they will do harm, but that they will do $4,000 worth of tokens quietly.
Claudeforce: Claude Becomes Slack’s Default Model
Salesforce and Anthropic announced Claudeforce on August 26, and the scope is larger than the usual model-partnership press release. Claude is now the default model for Slack AI and Slackbot, plus Headless 360, Agentforce Coworker, and Claude Code across Salesforce’s own engineering organization.
The traffic runs both directions. Salesforce in Claude is a plugin carrying 37 prebuilt sales skills that let sellers and agents reason over live revenue context, automate pipeline updates and take governed action without leaving Claude. Claudeforce is in pilot with select customers now, with open beta expected in September 2026 and additional prebuilt skills landing in Q3.
The Slack default is the piece that matters most, and it is easy to under-read. Slack is where the unstructured record of how a company actually works already lives — decisions, context, half-finished threads. Becoming the default reasoning layer over that corpus is a materially better position than being the model a company chose in a bake-off, because nobody re-evaluates the thing that is already answering their questions.
$45 Billion for 460 Megawatts Microsoft Didn’t Want
Anthropic agreed to spend $45 billion over six years leasing AI cloud capacity from Nscale’s flagship campus in Mason County, West Virginia — roughly 460 megawatts of power. Nscale will deploy Nvidia Vera Rubin silicon, with the facility expected to come online at the end of 2027.
The backstory is the story. Microsoft walked away from a 1.35-gigawatt commitment at this same campus weeks earlier. Anthropic stepped into roughly a third of it and, in doing so, became the anchor tenant underwriting Nscale’s IPO. Two companies preparing to go public, one of them de-risking the other’s prospectus.
Set the number next to the rest of the week and the shape of the bet is clear: Anthropic is signing six-year, tens-of-billions compute commitments against revenue that is growing fast but has not yet been tested through a downturn — while simultaneously fielding a copyright suit whose damages exposure is measured in billions. Both of those are IPO-window problems, and they point in opposite directions.
Anthropic Is Buying Distribution Faster Than It Is Settling Its Debts
Read this week’s stories in the order they will matter to an underwriter and a pattern falls out immediately.
On the asset side: Claude became the default model inside Slack, which is the single best distribution surface in enterprise software. Free Claude for Teachers went out to every US school district willing to sign, buying a generation of habit for the cost of inference. Enterprise-managed MCP auth, geo pinning and Admin API GA removed three of the standard procurement blockers in a single month. And $45 billion of compute got locked in ahead of demand. Every one of those is a move a company makes when it believes distribution, not capability, is the contested ground.
On the liability side, the Sony and Warner Chappell complaint is not the same class of problem as the suits that came before it. Naming Amodei and Mann personally is a pressure tactic, but the substantive risk is the output-substitution theory — the claim that Claude reproducing lyrics competes directly with the licensed lyrics market. Training-data arguments live in an unsettled corner of fair use. Market-substitution arguments do not; that is the fourth factor, and it is the one courts weigh most heavily. Anthropic operating under existing lyric guardrails cuts both ways here: it demonstrates good faith, and it demonstrates that the company already conceded the outputs were a problem worth constraining.
The through-line connecting the retention reversal to everything else is that Anthropic keeps making enterprise-legible concessions ahead of being forced to. It published that its own retention policy posed “real risks to our business success,” then changed it. That is not a safety company blinking. It is a company that has figured out its actual moat is being the frontier lab a general counsel will sign off on — and is spending capability leverage to buy legal legibility.
Which makes the timing of the music suit genuinely bad rather than merely inconvenient. Every other item on this list is Anthropic converting technical position into enterprise trust. A federal complaint alleging systematic piracy, with founders named, is the one input that runs directly against the thing being purchased. The compute is contracted, the distribution is signed, the compliance work is shipping. The unresolved question in the S-1 will not be whether Claude is good. It is what the training corpus cost, and whether that bill has fully arrived.