AI-noleak – Local secret proxy for AI CLIs
What they're not telling you: AI Models Are Leaking Your Secrets to Corporate Servers. The SEC Isn't Watching. Software developers are sending AWS credentials, database passwords, and SSH keys directly into the servers of Anthropic, OpenAI, and Google every single day—and the infrastructure to stop it exists, but remains marginal, unknown, and entirely unregulated.
What the Documents Show
That's the reality exposed by ai-noleak, an open-source project that does something the $150 billion AI infrastructure industry has resisted: it intercepts outbound prompts before they reach commercial AI APIs and strips out credentials, tokens, and API keys before they can be stored, logged, or trained on. The tool replaces exposed secrets with deterministic placeholders—@TOKEN_8f51a2@—then restores the real credential locally, ensuring developers get usable output without surrendering their infrastructure keys to third-party vendors. The technical elegance of the solution highlights a systemic problem: this layer of protection should have been built into Claude, ChatGPT, and Gemini from inception. Instead, Anthropic, OpenAI, and Google have left credential protection to individual developers operating ad-hoc tools in a fragmented ecosystem. The result is that millions of prompts containing authentication material flow daily through corporate infrastructure operated by companies whose business model depends on accumulating training data.
Follow the Money
Here's what the mainstream AI narrative obscures: credential leakage through AI interactions is not a hypothetical risk. It's a live, ongoing exfiltration vector. When a developer uses Claude to write an S3 upload script and accidentally includes their AWS access key ID in the prompt—something the ai-noleak documentation shows is routine—that credential now exists in Anthropic's request logs, potentially in training datasets, and almost certainly in the company's cold storage systems. Anthropic's own privacy documentation states that prompts and outputs may be retained for "safety, security, and fraud prevention." The definition of those terms remains opaque. The SEC has jurisdiction over the disclosure practices of these companies. Anthropic received $5 billion in Google investment, triggering securities law obligations.
What Else We Know
Yet no examiner at the SEC's Office of Compliance Inspections and Examinations (OCIE) has initiated enforcement action on credential handling standards. OpenAI, valued at $80 billion in secondary markets, remains largely unregulated as a private company, though its access to Federal Reserve banking infrastructure through Microsoft partnerships and its contracts with U.S. defense contractors should trigger examination authority. Neither company has disclosed the scope, frequency, or severity of credential leakage incidents to investors or the public. The ai-noleak project reveals what regulatory capture looks like in real-time: a clear and present vulnerability, a working technical solution, and complete institutional indifference from the agencies tasked with protecting financial systems and market participants. Developers at major financial institutions—JP Morgan, Goldman Sachs, Bank of America—are using ChatGPT and Claude for internal development tasks.
Primary Sources
- Source: Hacker News
- Category: Corporate Watchdog
- Cross-reference independently — don't take our word for it.
Disclosure: NewsAnarchist aggregates from public records, API feeds (Federal Register, CourtListener, MuckRock, Hacker News), and independent media. AI-assisted synthesis. Always verify primary sources linked above.