Money20/20 Key Takeaways: Building Trust in an Agent-Led Economy

Summary

• Money20/20 highlighted how quickly the conversation around AI agents, identity and trust is evolving.
• AI agents are creating new questions around identity and authority as they take on increasingly autonomous roles.
• Common protocols could bring greater consistency, but widespread adoption remains a challenge.
• Network and mobile validation could offer another layer of trust, helping tackle fraud across human and AI interactions.

Going into Money20/20, we wanted to have an underlying theme that we could use to tie together the broader discussions on the future of Financial Services and Insurance (FS&I). This year we landed on cybersecurity—hyper-relevant, with huge growth and interest in the AI-led world we are in today. The interactions coming from this both validated the demand and revealed more insights into the Chief Information Security Officer (CISO) mindset.

After every event, we like to take a step back and reflect on the key insights and takeaways. This one stood out for its broad mix of thought leadership, market perspectives and opportunities to connect. Our key takeaways are:

We Are Still Assuming Human Speed and Scale Way Too Much

In April 2026, Anthropic handed a restricted frontier model called Claude Mythos to twelve of the largest technology organisations on earth under Project Glasswing: AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, Palo Alto Networks, the Linux Foundation and others. Unlimited model credits. Full cooperation. The best remediation engineering on the planet.

In the first month, the program surfaced more than 10,000 high and critical severity vulnerabilities. By May 22, Anthropic had disclosed 1,596 vetted findings to maintainers across 281 projects.

Under the best possible conditions, the fix pipeline absorbed roughly six percent of what the find pipeline produced, and under one percent of total discovery volume. Several open-source maintainers asked Anthropic to slow the rate of disclosure, because a high-severity bug still takes about two weeks to patch properly.

The part that lands hardest is this: the constraint is no longer detection. You will shortly be able to find far more than you can act on. The constraint is decision and deployment—triage, change approval, testing, release. Buying more discovery without rebuilding that pipeline makes the backlog visible without making it smaller.

This Just Highlights That the Game Has Changed

The ecosystem business models we have all proliferated just mean there are so many attack surfaces and points in and out of any organisation. And if we’re honest, there was a lag there anyway. We do a lot of Third-Party Risk Management (TPRM) at Synechron, and we see the emergence of new models there all the time as a good example of this. Our partner landscapes are also changing all of the time, and keeping up with this now needs new working models.

As a panellist on our MoneyPot Stage session highlighted, "you have to stop measuring effort and start measuring exposure." The bottleneck moved, and nobody moved the operating model with it.

So, CIOs and CISOs need to make that shift now—and this is an operating model challenge as much as anything else.

The role of AI and humans in response needs to be better defined. The triage is work machines do well. Architecture, threat modelling and control design are work they do not.

Key Trends in Cybersecurity

The security challenges of agentic commerce, open ecosystems, personal agents and increasing threats mean that we are in a collision of forces that are raising profound challenges for modern FS&I businesses.

The proposed framework is: identity, authority, then permitted actions.

There's a clear issue with allowing the good agents and only blocking the bad. Methods currently mean both tend to fail, and there are scenarios where you block or allow both.

Agent protocols are starting to address the need for common standards, with initiatives such as the Personal Agent Protocol emerging alongside other approaches. The real test, however, will be adoption. Much like the early days of e-commerce, establishing a standard is one thing; getting an ecosystem to converge around it is another.

That challenge becomes even more complex when identity enters the equation. Sovereign identity for agents is a growing focus, but defining an agent’s goals and parameters only gets us so far. The critical question is how we verify the human intent and authority behind its actions. As agents gain greater autonomy, proving who they represent and what they are authorised to do becomes fundamental to trust.

There's a general concern that this issue becomes "friction" in business models that rely on open transactions. Machine to machine “down-the-rails” is therefore seen as a way to solve friction issues in some transactions or processes. In other words, if the printer knows it has run out of ink, how do you allow it to order new ink directly?

When pressed on these ideas, everyone still uses predictive techniques to assess if something has malicious intent, showing that we need to find new ways to gain more certainty and trust.

E-commerce is a useful frame for FS&I organisations to try and shift their posture overall to allow for other customer experiences beyond “transactions” to occur and find ways for personal agents or third-party agents to transact with them.

The network is also emerging as an important new security layer. Mobile carrier and SIM validation can provide frictionless signals to help detect fraud, including SIM-swap activity, but they only solve part of the trust problem.

New approaches from companies such as SLC Digital (who we had the pleasure of hosting a closed roundtable with) are pushing this further, focusing on how we validate human intent as interactions increasingly involve humans, AI agents, or both.

So, What?

The overall takeaway from all of this is that Synechron’s focus on creating more adaptivity in clients is the right one to have.

We want our clients to become agentic, safe and secure FS&I businesses of the future. Our accelerators like RAPID look to address a shift in the working model, using new theologies, to allow cybersecurity teams to move at speed and get to outcomes that work.

There’s likely no one solution or move that will be the definitive answer. You’ll need multiple models, with the ability to swap out, recut and adapt agents at will. You’ll need to be able to operate in highly resilient environments where there’s no over-dependency.

Ecosystem businesses will still need to operate on open model, but this proliferation of partner services and embedded products and experiences will continue to change the threat landscape on an ongoing basis. So, you will need to focus on increasing your ability to convert the identification of threats or opportunities into outcomes.

AI speed is necessary, but it will be fused with human oversight and control, and we will need to continue to improve explainability. True ownership of our agentic capabilities and our security postures will need to shift and move like never before.