About SAPIEN

SAPIEN was created to address a specific gap: no open methodology existed for continuously scoring how models drift across independent behavioral dimensions, with conformance requirements and operational runbooks for production deployment. Existing benchmarks test what models know or how well they follow instructions. None measured what happens when a model knows the right answer but abandons it because a human pushed back.

The framework emerged from real-world testing of production AI deployments in the Managed Service Provider (MSP) ecosystem, where AI tools are deployed to businesses that depend on them for critical operations. When an AI assistant helps manage a company's cybersecurity posture, sycophantic drift isn't an academic curiosity — it's a vulnerability that can be exploited.

SAPIEN treats sycophantic drift as a measurable syndrome — not a single behavior — decomposed into four independent dimensions that can be scored, tracked, and compared. Each dimension captures a distinct mechanism of behavioral degradation: escalating specificity, disappearing risk disclosures, epistemic retreat, and emotional substitution. The result is a framework that doesn't just detect drift — it tells you how the model is failing and where to intervene.

Creator & Lead Author

Founding Contributors

The SAPIEN Protocol benefits from the expertise and guidance of its founding contributors — practitioners and technologists who shape the framework's direction across regulatory, behavioral, and systems domains.

Acknowledgments

SAPIEN started as a vulnerability disclosure and a question nobody had a good answer for: how do you measure AI behavioral safety? These founding contributors saw it early and committed their expertise to building the answer. This protocol exists because they said yes.

History

SAPIEN was developed through hands-on testing of production AI deployments in the Managed Service Provider ecosystem, where AI tools are deployed to businesses that depend on them for critical operations. The framework is maintained independently as an open standard under CC BY 4.0.

Contributing

SAPIEN is an open framework that welcomes contributions from AI safety researchers, red teams, compliance professionals, and practitioners. Areas where contributions are particularly valuable:

• New test scenarios across underrepresented domains
• Independent validation runs and benchmark data
• Scenario quality review against the authoring rubric
• Implementations in new scoring infrastructure
• SAPIEN Improvement Proposals (SIPs) for framework extensions

Submit contributions via GitHub or contact contact@sapienframework.org .

Governance

The SAPIEN Framework follows an open governance model. Framework changes are managed through the SAPIEN Improvement Proposal (SIP) process defined in Section 16 of the specification. A formal Steering Group is being established to provide independent oversight of the framework's evolution.

Organizations and individuals interested in participating in governance should contact contact@sapienframework.org.

Corporate Sponsors

The SAPIEN Protocol is an independent open standard. Corporate sponsors provide funding, engineering resources, and real-world testing environments. Sponsorship does not grant editorial control over the protocol specification.

Synthreo, as Founding Sponsor, contributes researcher time and compute resources that fund the SAPIEN benchmark scoring runs. Sponsorship carries no editorial control: the framework, scenario corpus, and published scores are maintained independently by SAPIEN Labs LLC.

Interested in sponsoring the SAPIEN Protocol? Contact partnerships@sapienframework.org

Get Involved