About Hamming

Hamming is the security layer for voice and chat agents.

We attack your agents the way bad actors would, show you exactly what they fall for, and keep every fix tested on every release and every live call.

Trusted by teams building for

  • Banks & financial institutions
  • Hospitals & healthcare
  • High-growth voice AI teams
Backed byY CombinatorMischiefAI Grant

Trusted by teams where a wrong answer is a liability.

  • Luma Health
  • Netomi
  • Y Combinator
  • Maven AGI
  • Ellipsis Health

Why voice agents need a security layer.

Sumanyu Sharma, Hamming founder and CEO, on stage with a guest
Sumanyu Sharma (left), Hamming founder and CEO

A customer-facing agent is a new public endpoint. Anyone with a phone number can reach it, and it can verify identity, read an account, and act on what it hears. It is also probabilistic: speech recognition, LLMs, text-to-speech, telephony, and tools each add variability, and a change to any of them can change what the agent does under pressure. In banking and healthcare, that is not a quality problem. It is a security and compliance problem with a liability attached.

We started Hamming in 2024 with evaluation infrastructure for AI products and launched automated voice agent testing that August, when the teams building voice agents had no way to test them beyond calling the number themselves. So we built the first platform dedicated to testing voice agents, and have since analyzed millions of production calls across banking, healthcare, and customer support. At that volume, the failures that mattered most were not quality problems. They were the kind a motivated caller could find and use.

So we turned the machinery around. The harness that ran thousands of realistic calls against an agent could also attack it the way a bad actor would, at a volume no manual review can match, and show whether a fix held. That is what Hamming became, and why banks, health systems, and enterprises bring us in: to get to the failure first. Your agent is answering calls right now, and somebody is testing it. The question is whether it is us or them.

What we believe.

  • A test expires on every change.

    Every prompt, model, or routing change is a new agent, so the checks have to run on every change, not once a year. Of 100 CISOs surveyed by a16z in October 2026, 18% run most of their offensive testing continuously today and 52% expect to within a year.
  • Evidence beats assurance.

    Anyone can claim their system is safe. Evidence is a recording, a transcript, and a rerun that shows the fix held, which is what your auditors, your board, and your regulator will ask for.
  • The vendor running the agent should not grade it.

    Hamming is independent of every agent platform, model, and telephony provider.
  • Break it first.

    We break agents to make them safer, and what we find goes first to the team that owns the agent.

Built by engineers from Tesla, Meta, and Citizen.

We have run production AI where a wrong answer reaches a real person: the ML behind hundreds of millions a year in sales at Tesla, real-time safety alerts to millions of people at Citizen, and Messenger at Meta, where failures at that scale are silent. Hamming is that discipline pointed at voice agents: break them before a customer or an attacker does, and prove the fix held. 70% of the team are former founders.

Releases
Simple feature requests, shipped
Tokens. Recognized by OpenAI alongside Stripe, SoftBank, and Jane Street
Security

Your data
stays yours.

We do not sell customer data, and we do not train on it. It is used to provide the service to you, and for nothing else.

SOC 2 Type IICertified
HIPAABAA available
GDPRDPA available
Sign-onSSO with Okta, Microsoft Entra ID, and Google Workspace
AccessRole-based access control per workspace
IsolationSingle-tenant deployment with customer-managed encryption keys
AuditFull audit trail, exportable to your SIEM
RedactionConfigurable PII and PHI redaction

Security &
compliance.

Built for regulated environments where trust, privacy, and audit readiness are non-negotiable.

  • SOC 2 logo

    SOC 2 Type II

    We maintain SOC 2 Type II controls to support enterprise security requirements for data protection, access controls, and operational resilience.

    Read the SOC 2 announcement
  • HIPAA Compliant badge

    HIPAA (BAA available)

    Hamming supports HIPAA-aligned workflows for testing and monitoring voice agents that handle protected health information (PHI). We can sign a Business Associate Agreement (BAA).

    Learn about HIPAA voice agent testing

On stage and in print.

Building the future of trust,
backed by founders.

We're backed by founders, operators and executives who have built generational companies. Together, we're helping teams ship AI voice agents that customers can trust. Hamming tests, red-teams, and monitors voice and chat agents, so teams catch leaks, jailbreaks, and broken promises before customers do.

  • Lauren Farleigh

    Lauren Farleigh

    Co-Founder & General Partner @ Mischief

    Mischief
  • Nate Abbott

    Nate Abbott

    General Partner @ Mischief

    Mischief