How are teams actually securing AI agents?

I'm researching how practitioners building with LLMs handle security and compliance: prompt injection, tool access, data leakage, and audit requirements. Your answers inform product research for Reinward, an AI agent security and compliance gateway built in the UK.

No marketing follows. I'll share the aggregated findings with everyone who takes part.

~5 minutes · 12 questions · findings shared back

section/01 → identity

About you

So responses can be credibly attributed, or anonymized if you prefer.

Name [REQUIRED]

Role / job title [REQUIRED]

Company or industry [REQUIRED]

May I reference your input in my market research? [REQUIRED]

Yes, name and role are fine Only anonymized (e.g. "a senior platform engineer") No, keep my answers private

section/02 → current_state

Your current reality

What you're building and what already worries you.

Are you or your team building or deploying LLM based applications or agents? [REQUIRED]

Yes, in production Yes, prototyping Planning to No

Which security risks concern you most? [REQUIRED]

Prompt injection Data leakage Unauthorized tool or API access Compliance and audit requirements Model output safety None yet

Pick as many as apply.

Have you experienced or tested any of these issues in practice? [OPTIONAL]

What do you currently use to address them, if anything? [OPTIONAL]

section/03 → the_gap

The gap

Whether a dedicated layer would earn a place in your stack.

If a gateway sat between your AI agents and your tools and data, enforcing access policies, blocking injections, and logging everything for compliance, would that solve a real problem for you? [REQUIRED]

Definitely Maybe No

What would make a tool like this a must have rather than a nice to have? [OPTIONAL]

Roughly what monthly budget could a problem like this justify for your team? [REQUIRED]

£0, we would only use open source Under £100 £100 to £500 £500 to £2,000 £2,000+ Not my decision

Interested in early access or being a design partner? [REQUIRED]

Yes Maybe No

Email, for early access and the aggregated findings [OPTIONAL]

Your answers are stored for market research only. Attribution follows whatever you chose above. See our Privacy Policy.