Molt AI disclosed in August 2026 that it raised $1 million in a pre-seed round, though the New York startup declined to name its backers or discuss valuation. The company builds adversarial testing tools designed to expose vulnerabilities in AI agents before they reach production environments.
The timing matters. Companies are rushing to deploy autonomous AI systems capable of executing tasks with minimal human oversight, even as regulatory frameworks tighten around them. The EU AI Act's robustness and cybersecurity obligations take effect on August 2, 2026, according to the regulation's text. Meanwhile, standards like NIST's AI Risk Management Framework and ISO/IEC 42001 are beginning to shape how enterprises govern these systems.
Grading agents by what they do
Molt's Fisher platform takes an unusual approach to testing. Rather than analyzing conversation transcripts, it grades AI agents by their action traces: the tool calls they make and the state changes they trigger. The company says this produces reproducible evidence and generates confidence-labeled findings mapped to multiple frameworks, including OWASP's agentic security top 10, NIST AI RMF, and EU AI Act requirements. Compliance with regulations such as HIPAA is also built into the assessment structure.
The startup published a white paper in July detailing a research corpus of more than 50,000 multi-turn adversarial episodes spanning more than 500,000 conversation turns. A field guide released the same month claims 10,000 reproduced findings and 533 attack strategies evolved from seed scenarios. Test cases range from customer support agents attempting refund-policy workarounds to coding assistants with unsafe file access, RAG systems vulnerable to prompt injection, and database agents with questionable permissions.
Background in crypto and infrastructure
Walton Comer, Molt's chairman and CEO, co-founded XBTO, where he served as chief investment officer. His brother Thomson Comer, now CTO, previously worked at NVIDIA on AI systems infrastructure and holds a master's degree in artificial intelligence from Colorado State University. Greg Frank, the company's chief scientist, came from Microsoft and studied at Harvard.
The startup also lists Joe Haggenmiller as president, Chris Fontes as head of product, and Rebekah Keida as head of marketing. The company would not disclose headcount.
A modest raise in a competitive market
Molt's $1 million haul looks modest next to recent checks written to competitors. Neo emerged from stealth in July with $100 million from Andreessen Horowitz, Bessemer Venture Partners, Craft Ventures, and Merlin Ventures to secure AI agents, AI-enabled applications, and browser-based identities. Tolmo raised $22 million from Accel and Y Combinator in June for a fleet of AI security agents in production environments. General Analysis closed a $10 million seed led by Altos Ventures in April, with participation from 645 Ventures, Menlo Ventures, and Y Combinator.
RELAI raised $6.9 million in June for verifiable continual learning, combining a $5.4 million pre-seed led by .406 Ventures with an earlier $1.5 million round. Jetty raised more than $2 million in May from AQC Capital and Hidden Layers Capital. At the mature end of the spectrum, Torq raised a $140 million Series D at a $1.2 billion valuation in January for an AI SOC platform.
Whether Molt's leaner capitalization reflects strategy or necessity is unclear. Some early-stage founders argue that smaller raises force discipline and sharper product focus, though others counter that in markets moving this fast, underfunding can be fatal.
Beyond testing
Molt is working on a broader architecture it calls Deep Model Trust, intended to move from testing toward what the company describes as "provable, bounded authority and verifiable behavior" for agents. Its roadmap includes components the website marks as "in production": a secure personal agent, small models, policy agents, and an agentic software development lifecycle.
Fisher is available now, and the company is marketing assessment engagements. It has not disclosed customers.
