# Reflection AI — Advancing open-model safety and agent security | Our partnership with Gray Swan

- Company: Reflection AI (reflection.ai)
- Announced: 2026-11-18
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://reflection.ai/blog/advancing-open-model-safety-gray-swan
- Record: https://forck.live/items/38099-advancing-open-model-safety-and-agent-security-our-partnership-with-gray-swan
- Subject: Reflection Beam
- Models affected: Beam

Reflection AI says it partnered with Gray Swan to red-team its open model Beam, testing safeguards against adversarial attacks and prompt injection in agentic systems. The company describes two distinct workflows for responsible open-model development: finding harmful behaviors before deployment, and checking that a model stays secure once it runs as an agent. Gray Swan's automated system Shade put Beam under sustained adversarial pressure across cyber, biology, chemistry, mass persuasion and influence operations, and general harmful content, making up to 200 attempts per scenario. In the cyber domain, Gray Swan tested 100 text-only scenarios mapped to MITRE ATT&CK techniques, including 54 malicious and 46 dual-use cases, and separately checked whether Beam gave functionally effective help on offensive cyber tasks as a coding agent. Reflection says the results showed where Beam's safeguards held and where more work is needed, and warns that low per-attempt success can still leave working paths for persistent attackers. The company also co-sponsored Gray Swan's latest public IPI Black arena challenge, and ran its own internal prompt-injection evaluation documented in its Safety Technical Report.

## Evidence

Verbatim from https://reflection.ai/blog/advancing-open-model-safety-gray-swan:

> Open models allow researchers and developers to inspect, adapt, and deploy AI on their own infrastructure. As these models become more capable, assessing their safety means testing how they respond to harmful requests, resist adversarial manipulation, and respect boundaries while carrying out legitimate tasks. Alongside our own internal safety evaluations, we've partnered with Gray Swan for adversarial testing and model security
> 
> Evaluating the Risks of Open Models
> 
> Responsible model development requires two related but distinct workflows. First, proactively identifying harmful behaviors a model can be induced to produce so you can address them before deployment. Second, ensuring the model behaves securely when deployed in an agentic system.
> 
> Both matter for models intended to support sensitive security workflows, but they involve different threat models and require different evaluation methods.
> 
> Pre-Release Testing on Safeguards and Capability
> 
> One critical part of evaluating a model before release is understanding how it responds to harmful requests, particularly when an attacker deliberately attempts to circumvent its safeguards.

---

Record: https://forck.live/items/38099-advancing-open-model-safety-and-agent-security-our-partnership-with-gray-swan
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md
