Building the Future, Responsibly

Bloom Labs exists to push the absolute frontier of artificial intelligence — but raw capability is only half the mission. Every model we ship, from BabyMax 5.76 up through our flagship Homan 2.96, passes through the same uncompromising safety pipeline before it ever reaches a user. We believe the most advanced AI systems in the world should also be the most rigorously tested, and we've built Bloom Labs' entire research culture around that principle from day one.

Our safety team works hand-in-hand with the research team that builds our models, not downstream of it. That means alignment considerations, failure-mode analysis, and misuse resistance are baked into a model's architecture from its very first training run — not bolted on right before launch.

Our Commitments

• Dedicated alignment research, focused on keeping increasingly capable models steerable, honest, and predictable as they scale.
• Internal red-teaming on every model before release, with full-time researchers stress-testing for jailbreaks, harmful outputs, and unexpected edge-case behavior.
• A formal Responsible Scaling Policy that ties expanded model capability to correspondingly stronger safety evaluations and deployment safeguards.
• Independent third-party audits of our largest models, conducted by outside security and alignment researchers before any public launch.
• Regular transparency reports detailing benchmark results, known limitations, and incident disclosures — including the tricky stuff, like Homan 2.96's multiplication quirk.
• A firm user privacy commitment: we do not sell user data, and we design our systems to minimize what's retained in the first place.

Safety isn't a checkbox at Bloom Labs — it's an ongoing research program that grows alongside our models. Read more about our alignment research »