The Autonomous Adversary: How AI is changing cyber risk for executives and investors

€211 billion. That is what German companies lost to data theft, espionage, and sabotage in the past year, three-quarters of it driven by cyberattacks. (Bitkom, Wirtschaftsschutz 2026). 

Now AI is changing the equation again. According to insights shared by Paul Price, Founder of CodeWall, autonomous AI hacking agents can reduce the cost of an attempted organizational breach to roughly $15–20, operating at machine speed without fatigue or the limitations of human attackers.

This question shaped the Cybersecurity & AI Tech Talks at OMMAX in Munich, where OMMAX and Raymond James brought together investors, executives, and cybersecurity experts to discuss how AI is reshaping cyber risk and what leaders need to do about it. 

The key takeaway: cybersecurity is no longer just an IT issue. It is an enterprise value issue. Companies may know what can fail, but far fewer know who takes control when it does or whether the business can keep operating.

The discussion highlighted six actions leaders should take now.

Six actions leaders should take as AI changes the economics of cyber risk 

1. Translate cyber risk into business impact

Christian Riede framed cyber preparedness around three questions: What can fail? Who owns it and takes control? Can the business continue operating? Most companies answer the first well, with risk registers and technology in place. The second and third are where OMMAX’s cyber due diligence work most often finds gaps.

Sandra Rauch made the same point from the CIO’s seat. AI initiatives are usually judged on business value, while security is treated as an IT issue. When something fails, though, it is the whole company that carries the impact, not the CIO alone. She added that budget follows business value, not technical risk. An incident report only unlocks funding once someone has translated it into revenue at risk. 

What leaders should do now: Make cyber reporting speak the language of the business. For each critical system, quantify what an outage would cost in revenue, customer relationships, and operations. Use that as the basis for prioritization, ownership, and budget. Be explicit about accountability too: outsourcing a function to a cloud or software provider does not outsource responsibility for the business risk.

2. Plan for attackers that never sleep and cost almost nothing

Paul Price presented CodeWall’s research on autonomous hacking agents. By CodeWall’s account, these agents gained access to the internal AI platform of a major consulting firm within two hours. CodeWall also described an agent that, without being instructed to, wrote its own text-to-speech code and tried to socially engineer another company’s AI voice agent over many rounds. (The target held up.) These were speaker-reported demonstrations and have not been independently verified by OMMAX.

His argument was less about a “super hacker” than about economics. AI removes the limits that used to constrain attackers: time, patience, and rare skills. Japhet Wuensch drew the consequence for investors. When the cost of an attack falls this far, the target is no longer only the large enterprise. A 50-person company in a private equity portfolio becomes worth attacking too. Many of the tools and testing routines in use today were built for human-speed adversaries.

What leaders should do now: Move from annual, point-in-time pen tests to continuous testing of your entire estate, the way an AI-enabled attacker would approach it. For investors, this applies to the whole portfolio, including smaller companies, and not only the flagship assets.

3. Get the basics right, starting with knowing every door

The speakers returned repeatedly to fundamentals. Paul Price compared a large organization to a corridor with many doors, where nobody knows how many there are. Internal systems that were never meant to be public end up exposed to the internet, credentials sit in code, and access rights run deeper than anyone intended. 

Legacy systems and acquisitions make this harder. Sandra Rauch estimated that most of the vulnerabilities in a typical IT landscape sit in the legacy world, and Paul Price noted that AI can now analyze an undocumented legacy system in hours, work that once took a skilled specialist months. Christian Riede added that during M&A, the attack surface can grow faster than integration. Anja Konhäuser pointed to a third source: AI pilots and “hidden technology” going live across the company without central oversight. 

What leaders should do now: Build and maintain an accurate inventory of applications, infrastructure, external exposure, and privileged identities, and assign a named owner to each. Layer your defenses so that no single control is a single point of failure. Make sure employees know your cybersecurity and AI usage guidelines, and train them regularly. In transactions, extend cyber due diligence to cover post-deal integration and remediation, not only the target’s current vulnerabilities.

4. Govern AI agents with action controls, not just access controls

AI agents can read, write, trigger payments, and execute workflows. Christian Riede highlighted that traditional identity and access management governs where an agent can go, not what it does once there. His example: an email agent with legitimate Outlook access can still send a message that damages your most important customer relationship. Access control alone cannot prevent that. 

Business context provides some guardrails, but implicit knowledge and common sense are exactly what agents lack. And putting a human into every loop would defeat the purpose of agentic AI. The risk grows as agents connect to financial processes and operational technology on the shop floor. 

What leaders should do now: Apply three principles to every agentic use case: know the identity, limit the action, keep the kill switch. Each agent needs its own identity. Its permitted actions should be explicitly bounded, and it should be possible to revoke access and stop actions at several layers. Where no effective action control exists yet, treat that as a conscious risk acceptance decided at management level. 

5. Prove resilience under pressure and invest before the incident 

Backups, disaster recovery plans and incident playbooks are not proof of resilience. In OMMAX’s due diligence work, resilience is often assumed rather than tested, and incident response plans are rarely exercised. Christian Riede’s recommended sequence was to close the obvious exposures and name owners first, then prove recovery under pressure. After that, decide what gets fixed, what gets funded, and what is consciously accepted.

The panel also discussed timing. Anja Konhäuser noted that willingness to pay for security is typically far higher after an incident than before one, which is when it matters. Paul Price agreed from experience: clients often act on findings only after something has gone wrong.

What leaders should do now: Run realistic crisis exercises that test decision rights, escalation, restoration times, and whether critical business processes can continue while systems are down. Take your findings to the board in terms of business impact. For investors, build these questions into cyber due diligence and the first 100 days of value creation. 

6. Build the right cybersecurity skills and expertise 

Awareness alone is not enough to ensure cyber resilience. Japhet Wuensch emphasized that as AI changes the threat landscape, organizations need the right people and expertise to translate cybersecurity awareness into effective protection.

This means investing in specialized talent, continuously upskilling existing teams, and collaborating with external cybersecurity experts where internal capabilities are insufficient. As threats become more sophisticated and AI-driven attacks accelerate, building resilience requires not only the right technology and processes, but also the people who can operate, challenge, and continuously improve them.

What leaders should do now: Assess existing cybersecurity capabilities and identify critical skills gaps. Invest in hiring specialized talent, provide ongoing training for existing teams, and leverage external expertise where needed. Ensure cybersecurity capabilities evolve alongside AI adoption and changing business risks.

The takeaway

The speakers agreed that AI makes both innovation and attacks faster. Most of the defence still comes down to decisions leadership teams already know how to make: knowing what you own, assigning accountability, developing the right capabilities, testing what you rely on, and funding it before the incident. What has changed is the pace. The evening's closing message was that cybersecurity is a prerequisite for sustainable AI-driven transformation and long-term enterprise value, not a cost to be minimized.

Want to discuss what AI-enabled attackers mean for your organization or portfolio? Get in touch with our Cybersecurity & Tech DD experts. 

About OMMAX

OMMAX is a leading AI-first consultancy and AI-engineering platform specializing in AI strategy, business transformation, transaction advisory, and value creation in the age of AI. Founded in Munich in 2011, OMMAX serves large corporates, mid-sized companies, and private equity firms across Europe and the US, with more than 4,000 completed projects and a Net Promoter Score of 90.