Not all AI tools are built the same. In a wild west of available AI software, do they have the security controls expected for financial services?

As the AI boom continues, many financial services firms are eager to adopt brand new AI tools, but must also balance this with their obligations to keep their data secure.

For financial services firms, certain technology controls are expected and synonymous with the applications they use. Traditional financial software typically has these mature technology controls built in, but many emerging AI platforms prioritise features over security. As a result, the controls financial firms depend on rarely make it onto the roadmap of AI applications until much later in their development.

In this article, we break down the key technology controls every organisation should look for in AI tools, explaining why they matter and how to spot gaps that could put your data at risk. To make things even easier, we’ve also created a downloadable comparison table showing how leading AI platforms stack up against these controls.

What technology controls do financial services firms expect?

Privacy and confidentiality

One of the key controls in the AI space, much like in the early days of cloud, is the guarantee that your data remains yours and won’t be used in any other way without your explicit permission.

Encryption

Whilst data is typically encrypted, it’s always worth reading the fine print. Vendors should clearly state that your data is encrypted at rest (when stored) and in transit (during transmission). However, we have seen some AI vendors that do not guarantee this level of protection.

Data residency

Choosing where your data is stored or processed by an AI vendor is often not just desirable but required by regulators. Yet some AI vendors give you no choice and offer little transparency about where your data actually resides.

Audit/eDiscovery

The ability to audit what people in your firm input into an AI tool is surprisingly rare, yet it’s important for tracking how people are actually using the technology and ensuring it’s used correctly.

Single sign-in

This feature allows you to log in with your work username and password, rather than creating a separate account. While often seen as a “nice to have”, you would typically have a significant amount of security controls and oversight on your work logins, whereas separate AI tool logins can bypass those protections and even hide login events from your security team.

Multi-factor authentication (MFA)

MFA is one of the best protections to stop the most common attacks, yet in many AI tools, it is treated as a premium feature or isn’t enabled by default. Some AI software vendors haven’t even built MFA into their tools at all, leaving their AI applications as easy targets for attackers.

Controlling new features

AI evolves at an incredible pace, with new features often released straight into AI vendor applications without the option to disable these features or control their rollout to your users. Some of these features may be undesirable, especially if they change the nature of how other controls work, such as data sharing.

Sharing

Sharing capability can be useful, for example, to share the output of an AI prompt with a colleague, but many AI vendors provide little control over how data is shared. Without safeguards, unauthorised users may gain access to what you’ve shared. In fact, recently, there have been cases where shared prompts made their way into Google search results, exposing the entire output of prompts publicly.

Data upload

Uploading your own data is a must-have feature in AI, but many platforms give firms no way to control what data is being uploaded or to see what data has already been shared. If your policy restricts certain information from being shared, without this control, you’re relying entirely on individuals to follow your policy correctly.

Third-party apps

AI vendors often allow you to connect to third-party apps from within their own AI software, typically Dropbox, Box and Google Drive. Without controls in place, this activity is often hidden from monitoring by security teams, leaving users free to move data around and potentially take it out of the firm.

Which AI is the most compliant?

Often, AI vendors aren’t mature enough to provide documentation to allow you to assess what security controls they have in place, usually because those controls don’t exist. However, some of the more mature AI vendors have made significant headway. Our top two most compliant AI vendors are;

1. Microsoft 365 CoPilot

The strongest in our assessment is Microsoft 365 Copilot. Whilst it may fall short of some competitors in terms of capability, it stands out for its robust technology controls, making it a trusted choice for safe AI use in the Financial Services world.

2. OpenAI ChatGPT Enterprise Edition only

Second to Microsoft 365 CoPilot, this vendor only offers data residency controls limited to large regions, i.e. Europe rather than the UK, and its Audit/eDiscovery features require some non-insignificant setup to get working.

How does ChatGPT compare?

OpenAI’s ChatGPT is one of the most popular AI tools, particularly in financial services, but it comes in five flavours: Free, Plus, Pro, Teams and Enterprise. Each license offers different levels of security and controls.

You can download our full comparison of ChatGPT versions and Microsoft CoPilot technology controls here.

What about other AIs?

If you're concerned about other AI applications, get in touch with us and we can provide you with a more complete comparison, covering Perplexity, Cursor, Fireflies and others.

AI is an incredibly exciting yet constantly changing technology. Whilst there’s growing pressure to utilise AI and a wide range of tools to choose from, established technology controls exist for a reason. If the AI tool you are using, or planning to use, lacks these safeguards, proceed with extreme caution.