Anthropic hard-blocks Claude from outputting eight copyleft licenses including GPL-3 and Apache 2.0. A Reddit user documented the behavior on Sonnet 4.6, revealing a policy that lets MIT and BSD licenses pass freely.
Key facts
- 8 copyleft licenses blocked: MPL-2.0, LGPL-2.1, EPL-2.0, GPL-3, AGPL-3, Apache 2.0, SSPL, BSL.
- 5 permissive licenses not blocked: MIT, BSD-2, BSD-3, Unlicense, WTFPL.
- Tested on Claude Sonnet 4.6 with Medium reasoning.
- Reddit user shared 12 share links confirming each test result.
- Anthropic has not publicly commented on the filter.
A Reddit user discovered that Anthropic's Claude refuses to reproduce the full text of eight major open-source licenses, while permissively licensing ones like MIT and BSD pass through without issue. According to the Reddit post, the user tested Claude Sonnet 4.6 with Medium reasoning using a prompt that explicitly asked for verbatim output of each license.
The blocked licenses are: MPL-2.0, LGPL-2.1, EPL-2.0, GPL-3, AGPL-3, Apache 2.0, SSPL, and BSL. The unblocked licenses: WTFPL, MIT, Unlicense, BSD-2, and BSD-3. The user shared share links for each test, showing Claude refusing to reproduce the text with messages like "I'm not able to reproduce the full license text verbatim."
Key Takeaways
- Anthropic blocks Claude from outputting 8 copyleft licenses including GPL and Apache 2.0, while allowing MIT and BSD.
- A Reddit user documented the behavior on Sonnet 4.6.
Why Anthropic Draws the Line
The blocked licenses all contain copyleft provisions that restrict commercial use or require derivative works to carry the same license. Apache 2.0, despite being more permissive than GPL, includes a patent grant clause that could create legal exposure. SSPL and BSL are explicitly designed to prevent cloud providers from offering the software as a service without paying.
Anthropic's filter appears to be a risk-avoidance measure: reproducing a copyleft license verbatim could be interpreted as Anthropic distributing that license, potentially binding the company or its users to its terms. Permissive licenses like MIT and BSD impose no such obligations, so they pass through.
The move mirrors similar content filters at OpenAI and Google, which also restrict certain license outputs. However, Anthropic's list is notably broader — including Apache 2.0, which is widely used in enterprise software.
Implications for Developers
For developers using Claude to generate license headers or boilerplate, this filter creates a practical hurdle. A prompt asking Claude to "add an Apache 2.0 license to this project" will fail unless the developer manually pastes the text. The user who discovered the block noted that Claude's system reminders are "overzealous regex triggers, not actually intelligent, context-aware reminders."
Anthropic has not publicly commented on the filter. The company's safety-focused posture, built on Constitutional AI, may extend to legal liability avoidance. [Anthropic's prior statements] emphasize responsible AI use, and blocking copyleft licenses fits that narrative.
What to Watch
Watch for Anthropic to clarify the policy in a blog post or changelog, especially as Claude Code adoption grows and developers encounter the block mid-workflow. If competitors like OpenAI or Google adjust their own license filters in response, expect a broader industry alignment on which licenses are safe to reproduce.
Source: reddit.com
