Network and cybersecurity vendor Cloudflare on Thursday unveiled a tool suite that provides security and control for AI applications.
Cloudflare for AI helps users protect themselves against the threats of using AI models, including employees misusing the technology, toxic prompts and the leakage of personally identifiable information.
The platform is aimed at businesses, developers and content creators looking to adopt, deploy and secure AI technologies at scale safely.
The tools can help customers automatically discover and label AI applications so security teams can review and enact measures to safeguard usage. It also provides a gateway system that enables visibility into the different employees and teams using AI apps. This system prevents employees and users from leaking or submitting sensitive information, the vendor said.
AI everywhere and a need for secure AI
With this release, Cloudflare is responding to the growing trend of "AI everywhere" in businesses.
The rush to adopt generative AI systems including large language models (LLMs) and the proliferation in technologies like agentic AI also means "an inevitable expansion of an organization attack surface," said Rik Turner, an analyst at Omdia, a division of Informa TechTarget.
With regular machine learning as well as with LLMs, there is the risk that training data may be poisoned, thereby skewing the results of any analysis.
Rik TurnerAnalyst, Omdia
"With regular machine learning [ML] as well as with LLMs, there is the risk that training data may be poisoned, thereby skewing the results of any analysis," Turner said. "With GenAI, there is the risk of inappropriate usage, and with agentic AI, the risk that automated processes may get out of control and wreak havoc."
Cloudflare for AI addresses both traditional ML and LLM-based AI. The new suite also provides developers more tools for different stages of the AI lifecycle, said Fernando Montenegro, an analyst at Futurum Group.
If a developer or user needs governance tools, they're in the suite. If a user needs the AI system to be protected against scraping, the tools for doing that are also there.
"This is helpful," Montenegro said. While the set of features being implemented is not unique, the fact that developers can integrate them into their AI projects is beneficial, he added.
For one, it competes with vendors like HiddenLayer and Protect AI, which provide security for AI capabilities. It also vies with cloud giants AWS, Microsoft Azure and Google, which already have tools that address AI security.
AWS has Amazon Macie to discover and protect sensitive data and detect different data types. Google has a Secure AI Framework for building and deploying AI systems securely.
Another challenge for Cloudflare is that many chief information security officers still need help understanding the tools and services they can use to address their specific problems.
"They need to make sure that the tooling that they are making available … is effectively solving the problems that people have," Montenegro said. "They have to be fit for function."
So, Turner said Cloudflare needs to educate the market more about the security applications the new suite addresses.
"It would also be worth explaining whether its security for AI services is only for companies that have their websites hosted by Cloudflare already, or is it available to all and any organization, regardless of whether it is already a Cloudflare customer?" he said. "In other words, how heterogeneous is Cloudflare for AI?"
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence
