Control your IP
Know which text your models produced, and prove it. When your output turns up in a competitor's product, a scraped dataset or a court filing, the claim is verifiable — not a matter of opinion.
AI provenance & security
Security Objectives builds XAMMY, a watermarking and provenance runtime for large language models — built by security researchers who have spent twenty years finding the gaps other teams overlook.
The quarterly report shows revenue grew steadily across all regions, driven by strong demand in emerging markets.
Why it matters
Generated text is everywhere and, until now, anonymous. XAMMY gives the people and businesses who produce it a way to keep custody of it — and gives everyone else a way to check. For a writer, that means proving what you wrote and what you didn't. For a business, it means keeping control of what your models put into the world.
Know which text your models produced, and prove it. When your output turns up in a competitor's product, a scraped dataset or a court filing, the claim is verifiable — not a matter of opinion.
Readers, platforms and publishers can check what was written by a machine, even after it has been edited, quoted or paraphrased. Authenticity becomes something you verify, not something you take on trust.
Every generated output can carry a signed record of where it came from, and an audit trail that survives editing. When a regulator, a client or a court asks, you have the answer on file.
What we build
A vLLM-native watermark that survives paraphrase, with signed provenance records and a detection service. Runs on NVIDIA GPUs in your cloud or on-prem.
Learn more →
Verify authorship, measure watermark survival under attack, and export an audit trail your compliance team can hand to a regulator.
Learn more →
XAMMY ships with its own adversarial program: paraphrase and back-translation sweeps, LLM-rewrite attacks, and jailbreak harnesses. Results are published as they are produced.
Learn more →
Product
XAMMY — LLM watermarking & provenance
Runs on
NVIDIA CUDA (vLLM) · Apple Silicon (MLX)
Status
Working system, pre-release · onboarding design partners
Design partners
We are taking on a small number of design partners in publishing, financial services and legal.