How to innovate with AI without compromising security? Discover the keys to successful governance between companies and freelancers: risk management, tackling "Shadow AI," and AI Act compliance.
Innovating fast is one thing; innovating securely is another. Today, 78% of companies have already integrated AI into their processes, yet only 7% have implemented the necessary safeguards.
This gap presents a major challenge. On Malt, where innovation happens daily via thousands of projects, this imbalance presents a certain level of risk.. During a recent Malt Academy session, Adam D. Wisniewski, Tech Strategy Director (AI & Blockchain) and freelance expert on Malt, was blunt: without security, innovation is a ticking time bomb.
Because freelancing is the primary vehicle for AI adoption in companies, security is no longer just a technical option: it has become the foundation of the contractual relationship and the trust between independent experts and their clients. Here are the key takeaways from his session.
The first challenge of governance, as highlighted by Adam, is understanding that AI is not a classic kind of software.
A traditional program is deterministic: a precise input always produces the same result. AI, however, is probabilistic. It doesn't follow a fixed path, which fundamentally changes the game. As highlighted in our Malt Tech Trends 2026 report, this unpredictability requires us to rethink software structure around three zones of friction:
Classic software is a fixed structure. AI is "alive" through its data. If you don't master the training pipeline (its "diet"), you lose control. This leads to "Drift": over time, an AI may produce biased or dangerous results simply because real-world data has changed.
Sometimes, even the AI cannot explain its own choices. For a banker or an HR manager, this is a legal nightmare. The solution? XAI (Explainable AI): teaching AI to be accountable.
We are moving from AI "Copilots" (suggestion) to AI "Agents" (execution). An agent with the keys to your GitHub or emails is brilliant... until it gets hacked. At that point, it’s no longer virtual; your bank account takes the hit.
AI also “lives" through the dynamic, adaptive way in which people work with it. Changes in the way we ‘collaborate’ with AI can give rise to new risks. For example, if we delegate further responsibilities, incorporate additional data, or simply stop reviewing its outputs due to over-reliance.
The cyber threat landscape has grown more and more complex. During the session, Adam identified the primary technical risks to monitor closely:
This is the new face of SQL injection (a classic technique where a hacker hijacks a system by inserting malicious code). In short, it’s the art of "gaslighting" an AI…
If a freelance expert or a company trains a model (Fine-tuning) on public data without filtering it, an attacker may have "polluted" that data beforehand. The goal? To create a backdoor in the model that only activates during a specific scenario planned by the hacker. This is the 2026 version of the Trojan Horse.
Through targeted API queries, attackers can reconstruct sensitive data used to train the model (e.g., finding client names or trade secrets) simply by observing the AI's responses.
"The greatest danger is not the AI itself, but the illusion of control. Securing AI is not about building a wall around the model; it is about ensuring the integrity of every piece of data that enters and every decision that comes out.""
On an AI mission, your role goes beyond that of a technical expert: you become a trusted partner. According to Adam, your mission is to reassure clients by avoiding the trap of "Shadow AI” and to act as the first line of defense in the security architecture.
Today, using AI is no longer a question of legitimacy: clients actively seek out freelancers who integrate these tools to boost productivity. However, this expectation comes with a requirement for absolute security to avoid clandestine use of unsecured tools by staff or contractors.
To turn AI into a professional asset, the freelancer must be proactive:
Adam D. Wisniewski insists that security must be a core ingredient of your project, not just a cherry on top once it’s out of the oven.
For companies, the Malt Academy session outlined three pillars of security: technical, human, and legal.
The most sophisticated technique cannot stop human error. As they say in IT, the problem is often "between the chair and the computer."
History repeats itself: just as some employees previously sent ultra-confidential files to free PDF conversion sites without a second thought, the risk today is "pasting" the year's financial report into a public AI (ouch... bad idea). Training must become mandatory for all employees using these tools. A culture of "algorithmic vigilance" is no longer an option.
With the arrival of the AI Act in Europe, companies must now classify their AI systems by risk level. This is no longer just "best practice"; it is a legal obligation for documentation and transparency.
In this context, working with freelancers who master this regulatory maze (GDPR + AI Act) is essential. The expert ensures the tool is "compliant by design," protecting the company from fines that can reach up to 7% of global turnover.
Some see security as a brake. It is the opposite: it is the seatbelt that allows you to drive at 200 km/h.
Companies adopting secured AI could see productivity increase by 60%. Why? Because they will automate critical processes (Accounting, R&D, Legal) that their fearful competitors won't dare to touch.
In a world saturated with AI-generated content, trust is the number one differentiator. A company that proves the robustness of its AI will attract better talent and more clients. A freelancer who certifies their security processes can justify higher day rates (TJM) because they reduce the client’s operational risk.
AI presents a paradigm shift comparable to the arrival of the Internet. Security is not a "checkbox" on a form; it is a shared responsibility. As Adam D. Wisniewski summarized, the future belongs to those who marry creative power with cybersecurity rigor. At Malt, we don't just connect talent with projects; we create the trusted framework essential for this revolution. By curating experts who master these security challenges, we help our clients turn AI into a serene growth lever.
A freelancer who protects their client becomes an indispensable partner; a company that adopts these reflexes becomes a magnet for top talent.
Find the full technical analysis and attack/defense demos in the Malt Academy replay with Adam D. Wisniewski.
With 25 years of professional experience, a MSc in Theoretical Physics from the ETH Zurich and Executive Programs from Yale, Oxford and Harvard, Adam is helping companies to identify and reach their strategic goals harnessing the potential of new technologies, and especially AI.
Having consulted many large and mid-sized banks on risk management and compliance topics working for consulting companies like Accenture or BearingPoint, Adam knows how important it is to use technology safely. In this regard, AI presents us with new challenges that we must overcome in order to realize the full potential of AI. Adam currently applies his experience through his own consulting boutique AI4Leaders.
