Copyleaks Codeleaks AI source code detector

Code is metered in tokens. The page's FAQ reads: "Every 250 tokens count as 1 credit. 251-500 tokens = 2 credits, 501-750 tokens = 3 credits, and so on", and the 2024 capture gave the worked example that in a Java program "class, HelloToken, {, static, void, main, (, String, args, [, ], ), System, ., out, println would all count as tokens" (copyleaks.com/codeleaks and the 9 December 2024 capture, read 9 November 2026). A "simple, secure, locally run Python" Credit Calculator on the vendor's GitHub page forecasts credits for a scan. No file size limit is published, and no free allowance is stated for Codeleaks itself; the platform grants new users a limited number of free credits, with no count published.

The Copyleaks pricing page prices credits in words and images, not tokens, and shows the same two self-serve plans that the Copyleaks AI Detector card carries; Codeleaks is sold through the Enterprise tier, described there as "Enterprise-level detection for IP, code licensing, and governance through one integration" (copyleaks.com/pricing, read 9 November 2026). Figures below are as displayed in dollars on 9 November 2026.

Retention follows the Copyleaks Privacy Policy, last revised August 24, 2026, which states that Copyleaks may act as a controller over customer data "for purposes such as fraud detection, billing and improving Copyleaks' services including by training Copyleaks' models", that "If you are a direct customer, you may opt out of our use of your information to train our models by contacting customer support", and that the Hide Sensitive Data feature "is currently available for text-based Content only" (copyleaks.com/legal/privacypolicy, read 9 November 2026). The retention section keeps personal information "for as long as necessary for us to provide our services" and deletes an Enterprise-linked account "within 90 days after the expiration or termination of our agreement with the Enterprise Customer"; Enterprise Customer Data is processed under a Data Processing Addendum outside the policy. No period is stated for submitted code, so the retention rank field is empty. The Codeleaks page promises to "Prevent proprietary code from being stored in AI repositories", a sentence about other companies' models rather than Copyleaks' own storage.

The launch release is the only place the vendor puts a figure on code: Codeleaks "Detects AI-generated source code with 99.8% (accuracy)" (press release, November 9, 2023, read 9 November 2026). No sample, dataset or test date accompanies it, and the release predates the deprecation notices by 21 months.

The live Codeleaks page carries a second figure: "Detect genAI with over 99% (accuracy)*, a 0.2% false positive rate, with the ability to detect AI text paraphrased or interspersed with human text", footnoted "*Accuracy rating is based on internal testing of the English language datasets" (copyleaks.com/codeleaks, read 9 November 2026). The footnote and the word "text" place that sentence with the text detector, whose own methodology page the Copyleaks AI Detector card quotes at 0.30% false positives on the default setting; the card records it here because it appears on the code product's page, and keeps it out of the code figures. A blog post on third-party studies adds that "Copyleaks was selected as the most accurate and reliable AI content detection tool for source code, outperforming other options in four separate academic studies", without naming the four on the page (copyleaks.com/blog, read 9 November 2026).

The gap between the marketing and the documentation is the card's caveat. A page titled "AI Code Detection" and a release claiming 99.8% (accuracy) describe a capability that the same company's API documentation says it stopped returning for source code on December 16, 2025. A buyer who reads only the product page would reasonably expect an AI verdict on code; the release notes are where the vendor says otherwise.