Seattle-based startup Glacis is tackling what co-founder and former Microsoft Azure product leader Rohit Tatachar calls the next critical frontier in AI security: verifiable proof of agent behavior. The company's approach focuses not on model quality or training data integrity, but on creating cryptographic evidence that AI agents have operated within defined safety boundaries during execution.

Glacis emerged from stealth with a founding team that includes Tatachar as CTO, bringing significant enterprise cloud experience from his tenure at Microsoft where he led product development for Azure's security and management services. The startup's core proposition addresses a growing concern in enterprise AI adoption—how organizations can trust that autonomous agents haven't been compromised or manipulated during runtime.

The Proof Problem in AI Security

Traditional AI security has focused primarily on model vulnerabilities, training data poisoning, and prompt injection attacks. Glacis argues these approaches miss a fundamental challenge: even perfectly secure models can be subverted during execution if the runtime environment or agent interactions are compromised.

"We're seeing enterprises deploy AI agents for everything from customer service to financial analysis, but they have no way to prove these agents haven't been tampered with," explains a Glacis technical document. "Current monitoring solutions provide logs and alerts, but logs can be altered and alerts can be missed. What's needed is cryptographic proof that can stand up to audit and compliance requirements."

The problem becomes particularly acute in regulated industries like finance, healthcare, and government contracting, where organizations must demonstrate compliance with strict operational guidelines. Without verifiable proof of agent behavior, companies risk regulatory violations even when their AI systems appear to be functioning correctly.

Cryptographic Evidence for Agent Safeguards

Glacis's solution centers on creating tamper-proof cryptographic evidence that documents every action an AI agent takes during execution. The system operates by generating cryptographic signatures for each agent decision, interaction, and data access, creating an immutable chain of evidence that can be independently verified.

This approach differs fundamentally from traditional logging or monitoring systems. Where conventional security tools might record that an agent accessed certain data, Glacis creates cryptographic proof that the access occurred within authorized parameters and followed established safety protocols. The evidence chain includes timestamps, agent identifiers, action details, and environmental context—all cryptographically signed to prevent alteration.

"Think of it as a notary public for AI agents," says a Glacis technical overview. "Every significant action gets notarized with cryptographic proof that can be verified by third parties, including regulators, auditors, or business partners."

The system integrates with existing AI platforms and frameworks, operating as a middleware layer that intercepts agent actions, applies cryptographic verification, and generates the evidence chain without requiring significant changes to existing AI implementations.

Runtime Observability and Agentic Governance

Beyond simple proof generation, Glacis provides what the company calls "runtime observability"—real-time monitoring of agent behavior with the ability to detect anomalies and potential security breaches as they occur. This combines traditional monitoring with cryptographic verification, creating a dual-layer security approach.

Agentic governance represents another key component of the Glacis platform. The system allows organizations to define precise governance rules for AI agents, including what actions they can take, what data they can access, and what decisions they can make autonomously versus those requiring human approval. These governance rules are then enforced through the cryptographic proof system, with any violations immediately detectable and provable.

This governance framework addresses one of the most persistent challenges in enterprise AI deployment: maintaining control over increasingly autonomous systems. By combining predefined rules with cryptographic verification, organizations can grant AI agents greater autonomy while maintaining provable oversight.

Enterprise Applications and Use Cases

Financial services represent a primary target market for Glacis's technology. Banks and investment firms increasingly use AI agents for fraud detection, trading analysis, and customer service, but face stringent regulatory requirements around system integrity and auditability. Cryptographic proof of agent behavior could help these organizations demonstrate compliance with regulations like FINRA's supervision rules or the SEC's cybersecurity requirements.

Healthcare presents another compelling use case. AI agents in medical settings must comply with HIPAA regulations while making potentially life-altering decisions about patient care. Cryptographic evidence that these agents operated within approved protocols could help healthcare providers meet both regulatory requirements and malpractice insurance standards.

Government and defense applications also stand to benefit significantly. Federal agencies deploying AI for everything from benefits processing to intelligence analysis need verifiable proof that systems haven't been compromised by foreign actors or internal threats. The immutable nature of cryptographic evidence makes it particularly valuable for national security applications.

Even in less regulated industries, Glacis's technology addresses growing concerns about AI liability. As companies face increasing legal exposure for AI-driven decisions, having cryptographic proof of proper operation could become a valuable defense against lawsuits and regulatory actions.

Technical Implementation Challenges

Implementing cryptographic proof for AI agents presents several technical challenges that Glacis must overcome. The most significant is performance impact—adding cryptographic operations to every agent action could slow down AI systems, particularly those requiring real-time responses.

Glacis addresses this through optimized cryptographic algorithms and selective proof generation. Rather than applying full cryptographic verification to every minor action, the system uses a tiered approach where critical decisions receive comprehensive proof while routine operations get lighter verification. This balance between security and performance will be crucial for enterprise adoption.

Another challenge involves integration with diverse AI platforms and frameworks. The AI ecosystem includes numerous development tools, deployment platforms, and runtime environments, each with its own architecture and APIs. Glacis must provide flexible integration options that work across this fragmented landscape without requiring extensive customization for each platform.

Standardization represents a longer-term challenge. For cryptographic proof to achieve widespread adoption, industry standards will likely be needed around proof formats, verification protocols, and interoperability between different proof systems. Glacis's success may depend partly on its ability to influence or adopt emerging standards in this space.

Competitive Landscape and Market Position

Glacis enters a competitive but still-emerging market for AI security solutions. Traditional cybersecurity companies have begun adding AI-focused features to their existing products, while specialized AI security startups have emerged focusing on model security, data protection, and threat detection.

What distinguishes Glacis is its specific focus on cryptographic proof for agent behavior. While other companies might monitor AI systems or protect AI models, Glacis aims to provide verifiable evidence that can stand up to legal and regulatory scrutiny. This positions the company in a potentially valuable niche as AI adoption increases regulatory attention.

Tatachar's Microsoft background gives Glacis credibility in enterprise markets, particularly among Azure customers familiar with Microsoft's security approach. The startup could leverage this connection to build early adoption within the Microsoft ecosystem before expanding to other platforms.

Future Development and Industry Impact

The long-term success of Glacis's approach may depend on broader industry trends in AI governance and regulation. As governments worldwide develop AI regulations—from the EU's AI Act to proposed U.S. legislation—requirements for transparency and accountability will likely increase. Cryptographic proof systems could become essential for compliance with these emerging regulations.

Glacis also faces the challenge of educating the market about a relatively new concept in AI security. Most organizations currently focus on preventing AI security breaches rather than proving they didn't occur. Shifting this mindset will require demonstrating both the regulatory necessity and business value of cryptographic evidence.

Looking ahead, Glacis's technology could evolve beyond simple proof generation toward more sophisticated forms of AI accountability. Future developments might include automated compliance reporting, real-time regulatory alignment checking, or even blockchain-based proof storage for maximum immutability.

The startup's success could also spur broader innovation in AI security, encouraging other companies to develop complementary technologies or competing proof systems. This competition would benefit enterprises by providing more options and potentially driving down costs for AI security solutions.

Ultimately, Glacis represents an important recognition that AI security must evolve beyond prevention to include verification. As AI systems take on more critical roles in business and society, the ability to prove they've operated correctly becomes as important as preventing them from operating incorrectly. Whether through Glacis's specific approach or competing solutions, cryptographic proof for AI agents seems destined to become a standard component of enterprise AI infrastructure in the coming years.