AI-powered cyberattacks raise crypto exchange concerns
“Indian authorities have warned that hackers are using AI to target cryptocurrency exchanges and financial institutions, allowing attackers to identify vulnerabilities and execute attacks much faster than traditional methods.
“The speed of these attacks is particularly concerning. AI can automate vulnerability testing and help attackers adapt their techniques, giving security teams less time to identify suspicious activity and respond.
“I think the industry needs to focus on security beyond individual wallets. Institutions can have sophisticated custody arrangements and still be exposed through compromised credentials, weak internal permissions or vulnerabilities in connected applications.
“As institutional transaction volumes grow, security needs to cover everything from identity verification and transaction authorization to monitoring and settlement. Financial institutions need confidence that these systems can operate securely, particularly as attacks become more sophisticated.”
– Ryan Kirkley, co-founder and CEO, Global Settlement Network
AI’s growing influence
“Today, most companies are comfortable letting AI make recommendations. But by the end of 2027, I expect to move into actual decisions in areas where the rules are clear and the evidence is complete. That could mean validating an invoice against legal and mandate requirements, stopping a non-compliant document before it gets posted, or resolving routine mismatches between what was reported and what was booked.
“With tax determination on standard, high-volume transactions will likely follow once companies have enough confidence that every decision is logged and can be explained later. The important part is that autonomy won’t mean giving AI a blank check. Every decision needs a record of why it was made, clear boundaries around what the system can do, and someone accountable for those boundaries. When the rules are unclear or judgment is required, people still need to make the call.”
– Markus Hornburg, SVP, Global Compliance, Basware
Ambiguity will impede accountability
“Much of the conversation around AI currently centers on accountability — from formal agreements to new legislation. We’ll continue to see more of these pledges form in the new year. However, whether these groups or laws produce meaningful change is up in the air.”
“Terms such as ‘knowingly’ and ‘recklessly’ can make way for a heated debate, especially given how non-deterministic AI is in general. Ambiguous language like this will only cause more confusion among businesses of all sizes around who is responsible for AI-related security incidents.
“We need clear, concise language that defines the levels of severity and intentionality. For example, are there different consequences for incidents that occur from AI extensions that we authorize on our behalf, vs separate AI agent identities? Otherwise, we can expect to see inconsistent enforcement and leave us wondering, “What does it mean to be on the hook? What do they actually have to prove?””
– Matt Hillary, CISO and SVP, Security, Drata
2027 could bring an AI incident big enough to force a reckoning
“I don’t expect Congress to enact a comprehensive federal AI law in the next six months, and barring a major event between now and the spring, I don’t anticipate any real alignment on AI policy. There are too many competing government, regulatory and market interests.”
“Weak governance and controls over how AI systems are used can increase the risk of incidents. A serious incident could become a wake-up call for clearer guardrails, though it wouldn’t guarantee agreement on what those guardrails should be.”
– Shane Tierney, senior program manager, GRC, Drata
Agentic banking: AI needs certainty before autonomy
“2026 brought a lot of conversation about AI agents that don’t just answer financial questions but actually carry out operations: initiating payouts, moving treasury balances, paying suppliers. Adoption is still far behind the hype. Most of what we saw were pilots and controlled use cases, with a person still approving the final step.”
“In 2027, the question that decides whether agents scale won’t be capability; it will be liability. When an agent sends funds to the wrong account, or to a fraudulent one, someone has to answer for it: the business that deployed the agent, the company that built it, or the bank that executed the instruction.”
“Today there is no clear answer, and that will hold adoption back more than any limitation in the models themselves. Language models are probabilistic by nature, and money movement can’t be. An AI agent cannot operate without deterministic trust, meaning a verifiable, auditable verdict recorded before execution that confirms the account exists, belongs to who it claims, and that the operation is compliant. As more than 80% of B2B operations move toward machine-to-machine, the institutions that lead will be the ones that can prove, transaction by transaction, why an agent was allowed to move the money.”
– Ximena Alemán, co-founder and CEO, Prometeo
