ControlAI

Mitigating AI risk

Product

AI models such as ChatGPT have undergone multiple improvements to prevent cases of hallucination and misuse by bad actors, e.g. guarding against exploratory attacks aimed at revealing personal information. However, addressing these issues is not as simple as implementing rule-based solutions, rather it highlights an inherent vulnerability in current models.

Adversarial attacks exploit the way that machine learning picks up on patterns in data. Similar to other cybersecurity practices, this requires constant monitoring as users get more creative in their attacks.

Larger organisations have dedicated red teams which act as adversaries and try to break the models, however replicating certain threats is computationally expensive and time-consuming. Smaller firms are left dealing with traditional pen testing companies to understand the complex nature of machine learning models.

ControlAI is a SaaS platform that provides machine learning models with thorough robustness testing and automatically improves them.

Testing - Detailed analysis of vulnerabilities and potential attack vectors in the model

Benchmark - Comparison of the model's performance against state-of-the-art robustness and accuracy benchmarks

Improvement - Automatic improvement of the model to minimise the attack surface

Founder

Lukasz Bartoszcze, Founder & CEO - Previously deployment strategist at Palantir and consultant at BCG. PhD - Computer Science, University of Warwick

Spotlight Analysis

Strengths

  • Founder/market fit: with a PhD focused on ML robustness + Palantir/BCG, this is the ideal founder for this market

  • Strong tailwind: as more enterprises look to deploy machine learning models, there is a healthy demand for MLOps tools

Weaknesses

  • Scalability: it is challenging to scale with strong unit economics whilst having a tailored user experience

  • Crowded market: several traditional penetration testing players and well-funded new entrants operate in the space

Market

Robust Intelligence - Raised $30M seed round from Tiger Global Management and others in December 2021

Hidden Layer - Raised $6m seed round from Ten Eleven Ventures and others in July 2022

Why ControlAI?

ControlAI's unmatched domain expertise sets them up for success, whether that is through innovative testing methods, or by establishing a superior go-to-market strategy to capture a specific customer niche.

Key Details

Year founded: 2023

Summary: ControlAI is a SaaS platform that provides machine learning models with thorough robustness testing and automatically improves them.

Currently raising: £1m in the next 1-2 months

If you would like an intro to the founder, please reply to this email.

Note: if this email ended up in promotions/spam, make sure to move it to the primary folder if you are using Gmail, or check out these instructions if you use other mail providers.