People search Anthropic data breach timeline because the brand sits on billions of accounts, credentials, and cloud workloads. BreachHistory indexes 4 Anthropic-linked incidents, with headline counts up to 3K+ in catalog rows. This page maps every attested event through 2026 with internal links to canonical records.
Why Anthropic breach history matters
Anthropic operates in AI (United States). Across indexed rows, recurring themes include credential theft and social engineering, cloud and database misconfiguration, unverified actor or scraping claims. Understanding the chronological pattern helps security teams, customers, and regulators separate confirmed disclosures from forum marketing.
Full timeline through 2026
2026 — STDIO transport / SDK design; ecosystem-wide RCE risk (Ox Security; Apr)
Cataloged incident. In April 2026, Ox Security and follow-on trade press (The Register, The Hacker News, SecurityWeek, and others) publicly documented a systemic risk in Anthropic’s Model Context Protocol (MCP) ecosystem centered on STDIO-based transports: MCP SDK patterns that spawn local subprocesses from configuration can execute attacker-supplied operating-system commands with effectively no authentication boundary at the protocol layer, enabling unauthenticated remote code execution and full host compromise when combined with mal Exposed categories include Not a classic PII row count—risk is arbitrary code execution, secrets on host, and lateral movement in agent pipelines. No attested victim count is published for this row yet. See the anthropic-mcp-stdio 2026 record and canonical BreachHistory entry.
2026 — CLI source exposed via npm source map (~512k LOC TypeScript)
Cataloged incident. On March 31, 2026, developers and security analysts widely reported that the Claude Code CLI package on npm shipped with a published JavaScript source map (.map) that reconstructed a very large share of the product’s TypeScript source—on the order of ~512,000 lines across ~1,900 files in public summaries. The issue is generally described as an accidental publish (e.g., build tooling such as Bun emitting maps that were not stripped from the npm artifact) rather than unauthorized access to Anthropic’s customer databa Exposed categories include Proprietary TypeScript CLI/client source (via source map); not a traditional consumer personal-data breach. No attested victim count is published for this row yet. See the anthropic-claude-code 2026 record and canonical BreachHistory entry.
2026 — ~3,000 unpublished CMS assets exposed (Sanity misconfiguration; “Claude Mythos” drafts)
Unverified claim — treat actor counts cautiously. In late March 2026, security researchers publicly reported that Anthropic’s Sanity-based content management workflow had been misconfigured so that a large set of uploaded assets—on the order of nearly 3,000 unpublished files—were reachable without proper access controls. The exposure was accidental, not an intentional product launch. Materials described in Fortune and follow-on analysis included draft posts, PDFs, images, and internal documentation; some drafts referenced an unreleased model tier referred to in ma Exposed categories include Unpublished drafts, internal documents, PDFs, images—not end-user account databases; no traditional consumer PII count. BreachHistory cites approximately 3K+ affected records in this row. See the anthropic 2026 record and canonical BreachHistory entry.
2024 — — Contractor data
Cataloged incident. Phishing. Contractor data. Exposed categories include Names, emails, addresses, and other PII. No attested victim count is published for this row yet. See the anthropic2024 and canonical BreachHistory entry.
Patterns and analysis
- Credential theft and social engineering — appears across multiple Anthropic catalog entries; prioritize controls that address this class of failure.
- Cloud and database misconfiguration — appears across multiple Anthropic catalog entries; prioritize controls that address this class of failure.
- Unverified actor or scraping claims — appears across multiple Anthropic catalog entries; prioritize controls that address this class of failure.
- Record-count hygiene — BreachHistory indexes actor-cited figures separately from company-confirmed totals; read each row's technicalWriteup before treating counts as fact.
- 2026 monitoring — New disclosures roll into this timeline as they are verified or labeled unverified per catalog policy.
What to do if you may be affected
- Step 1: Enable phishing-resistant MFA on every account tied to this brand.
- Step 2: Use unique passwords and a password manager—breach rows often involve credential reuse.
- Step 3: Monitor official company breach notices and regulator filings, not dark-web downloads.
- Step 4: Bookmark the Anthropic company page for new 2026+ disclosures.
Canonical BreachHistory hub
Explore every indexed row: breachhistory.com/anthropic · Latest: anthropic-mcp-stdio2026.
Sources: BreachHistory catalog (4 rows for Anthropic), company and regulator disclosures cited in individual breach records.