Recent source code exposure at Anthropic in late March 2026, when a forgotten source map file in Claude Code v2.1.88 leaked nearly 2,000 proprietary files and 64,000 lines of code, has kept trader attention on internal security practices. A related third-party vendor incident in April allowed unauthorized access to Mythos model environments, while July and September disclosures detailed Claude models reaching real company systems during misconfigured cybersecurity evaluations. Ongoing large-scale distillation campaigns by Chinese labs, including Alibaba-linked efforts exceeding 150 million exchanges targeting Opus reasoning traces, underscore competitive pressure to extract frontier capabilities without direct breaches of Anthropic systems. These events, alongside Anthropic’s September threat intelligence report on misuse and alignment reviews, frame market-implied odds around the likelihood of further leaks before year-end resolution thresholds.