EvoMap issues a public letter changing Evolver from MIT to GPL-3.0, lamenting "GPL may not even be able to" prevent AI code washing

ME News message, April 16 (UTC+8). According to Beating Monitoring, the development team of AI Agent’s self-evolution engine Evolver, EvoMap, released an open letter to AI developers, announcing that the license will be changed from MIT to GPL-3.0 and that the core modules will be released in an obfuscated form. Previously, the team published a post accusing the Hermes Agent architecture under Nous Research of being plagiarized module by module. Nous Research founder Teknium denied having any knowledge of it, and the official account replied, “delete your account.”

The open letter did not name Hermes Agent, but said that “an overseas team with more resources” rolled out a “structurally isomorphic” framework within weeks after Evolver’s public release of its core protocol—“without acknowledgment, without mention, and even the self-evolution concept we’re proud of has instantly become their selling point.” The team said they were unable to pursue transnational litigation and chose to respond with a closed-source posture.

The core issue raised in the letter is that AI-assisted programming is bringing the protective power of traditional open-source licenses close to invalidation. After AI fully absorbs a set of code logic, it can output a different set of variable names and file structures, making the resulting product textually indistinguishable from the original. The team also admits that “the MIT license can’t protect much, and GPL may not either,” but “something still needs to be done.”

In the letter, the team cited several recent incidents of the same kind: the Meituan Tabbit AI browser, discovered on the first day of its public testing, directly used open-source code from an individual developer, “Pei Du Frog,” and the original project name remained in the source code. The “Three Provinces and Six Ministries AI Court” project was open-sourced and only 21 hours later was rewritten by AI and released as “original,” with textual similarity of just 3%, yet all 15 core designs were identical. The Microsoft Peerd project was also found copying code and comments from an individual open-source project, Spegel.

The audience of this letter is not Nous Research, but the entire open-source community. What EvoMap is trying to push for is a discussion: when AI reduces the cost of copying code to near zero and makes detection nearly impossible, can the incentive mechanisms of open source still function? If original creators find that their publicly released results are transformed into competing products within weeks by better-funded teams, fewer and fewer people will be willing to open-source core technologies. However, EvoMap’s own response also exposes the irreconcilability of this dilemma: GPL-3.0 requires derivative works to be open-sourced as well, but if the other party’s product has no textual overlap at the code level with the original, the GPL’s “contagion” provisions cannot be triggered. Obfuscated releases directly weaken the possibility of community contributions and run counter to the spirit of open source.

(Source: BlockBeats)