Cinder is a foundational player in the agent ecosystem, specifically occupying the safety and compliance layer of the stack. They use AI agents internally to automate decision-making workflows, such as traversing user graphs to find context for a moderation decision. More importantly, they provide the safety infrastructure that allows other agent-heavy companies to operate.
For developers building autonomous agents, Cinder is a model for how to implement 'guardrails' that are operational rather than just theoretical. They champion the idea of 'Trust & Safety Orchestration,' where the safety layer is integrated directly into the product lifecycle—from red teaming a model's weights to managing the real-time actions of an agentic interface. This makes them a critical partner for any lab shipping models or agents that interact with the public.
Trust and safety in the pre-AI era was largely a game of volume. Companies hired thousands of moderators to review flagged content against a list of static rules. Generative AI changed this math by creating an explosion of content that legacy systems cannot process. Cinder is a response to this shift, building what it describes as the infrastructure for trust and safety operations. Instead of treating content moderation as a support ticket problem, Cinder treats it as a security problem, applying principles from threat intelligence and national security to digital platforms.
Cinder is not just a collection of classifiers. While it integrates with tools like OpenAI’s moderation API, its value lies in orchestration. The platform functions as a command center where non-technical users can configure complex workflows. This includes real-time detection, automated policy enforcement, and human-in-the-loop triage. A key differentiator is how the platform handles context. Legacy moderation tools often evaluate a single post in isolation. Cinder builds a graph of user behavior, linking accounts, activities, and historical data to help its AI agents make informed decisions. This approach allows companies to catch repeat offenders and coordinated abuse campaigns that simple text or image scanners would miss.
Cinder was founded in 2021 by a team with deep roots in institutional safety. Glen Wise, the CEO, previously led security engineering for Meta’s community threat intelligence platform. He is joined by Philip Brennan and Brian Fishman, both of whom have backgrounds in US government counterterrorism. This pedigree is reflected in the product’s architecture, which mirrors the investigative workflows used by intelligence agencies to track bad actors. The company is based in Washington, D.C., and recently raised $41 million in Series B funding led by Radical Ventures to expand its reach among the labs building the next generation of foundation models.
High-profile AI labs like Black Forest Labs and Character.AI use Cinder to manage the unique risks of generative media. For Black Forest Labs, Cinder provided adversarial red teaming to stress-test the Flux.2 model before release, catching edge cases in refusal behavior and guardrail failures. For Character.AI, the platform helped reduce the burden on human moderators by 50% through automated orchestration. By moving from manual review queues to 'policy-aligned agents,' Cinder allows these platforms to scale without a linear increase in headcount. The company competes with generalist CRMs and specialized moderation startups like Hive or Unitary, but its primary target is the internal 'build' team that is tired of maintaining fragile in-house safety tooling.
A single command center for trust and safety operations and AI-driven content moderation.
Cinder is hiring