Leaked document reveals ‘Maverick,’ a new Amazon AI security initiative led by Uber’s former CISO

• Amazon’s security team has launched a new AI program called “Maverick.”
• The goal is to build automation tools for AI security and foster partnerships across the company.
• Amazon is doubling down on the generative AI space.

Amazon has launched a new internal security program devoted to generative artificial intelligence, as the retail behemoth continues to bet big on the still-in-development technology.

According to an internal document obtained by Insider, the new initiative, dubbed “Maverick,” focuses on developing new tools and fostering collaboration across Amazon to address the security and risk of Generative AI and large language models.

The project is led by four executives, including John “Four” Flynn, VP and CIO of Amazon’s retail business. Flynn was Uber’s CISO until 2020, when he left to join Amazon.

“Amazon Security launched the Maverick program in Q2 2023 to ensure LLM-based experiences meet our high security standards to protect customer data and trust.” According to the document, “Maverick’s approach is to partner with teams exploring GenAl opportunities, centralize and share security best practices, and be the Center of Excellence for developing GenAI securely at the scale of Amazon.”

Amazon’s spokesperson, Adam Montgomery, told Insider that the company has a “deep commitment to the safe, secure, and responsible development and deployment of generative AI.”

“Our company-wide approach to securing AI development paves the way for a future that maximizes the benefits of AI while minimizing risks,” according to the statement.

Maverick is part of Amazon’s growing effort to establish a stronger presence in the field of generative AI. Despite working in AI technologies for decades, Amazon has yet to capitalize on the popularity of GenAI like competitors Microsoft, OpenAI, and Google. The latest Llama 2 model from Meta is also making waves in the industry.

Amazon has asked almost every team within the company to come up with ideas that leverage generative AI and ChatGPT-like chatbot technologies. As previously reported by Insider, CEO Andy Jassy recently promoted an Alexa executive to lead a new group working on the company’s “most ambitious” language models.

Insider previously reported that the Amazon Web Services cloud unit established an independent team focused on assisting customers with the use of generative AI. In 2022, AWS will release new AI products such as Bedrock, a service that makes other foundational models accessible, and Code Whisperer, an AI coding tool.

According to the internal document, the new Maverick initiative wants to “understand GenAI risks” for Amazon and create “centralized GenAI security testing tools.”

Maverick is concentrating his efforts in two areas. The first component is tooling to “automate secure development, reviews, and testing” of GenAI models and applications. The second step entails collaborating with security teams to “create GenAI security guidance and add specific tasks to existing security mechanisms,” according to the document.

“We are consolidating risks, developing guidance, and creating tools to enable security,” the company stated.

LLMs are a key underlying technology for AI services. These models are trained on massive datasets in order to produce human-like responses. OpenAI, Google, and Meta, among others, have built massive, powerful LLMs that have wowed the tech world. Amazon has its own as well, such as Alexa Teacher Model and Titan.

Earlier this month, during Amazon’s earnings call, Jassy told analysts that AI is a top company priority and a “significant investment and focus for us.”

“On the AI question,” Jassy said during the call, “I would tell you that every single one of our businesses within Amazon has multiple generative AI initiatives going on right now.”