Harriet Farlow Calls for Defining Acceptable Risk for AI

Harriet Farlow, CEO of the consulting firm Mileva Security Labs, calls on information security leaders to define the risk they are willing to accept for their AI usage. She advocates for appropriate safeguards rather than a quest for infallible rules.
Leaders Must Be Able to Halt Projects
Agents are adopted without prior examination of the resources they will have access to, claims Harriet Farlow, who says she frequently encounters this situation. She also observes that some organizations dive into AI before assessing its security. This haste is notable, especially since most professionals involved have already experienced technological evolutions that were detrimental to security.
For her, the choice between a Chief AI Officer as the sole decision-maker and distributed responsibilities does not have a universal answer. The key lies in the visibility of leaders over each point of contact in the system and their ability to halt a risky project. Governance, in her view, should frame a progression at an appropriate pace, rather than prevent any advancement.
She also advocates for open exchanges with IT teams, security specialists, AI developers, and vendors involved in deployments. Tools from providers must undergo a risk assessment, she emphasizes, as organizations have less control over them.
Excessive Restrictions Can Encourage Workarounds
A too-restrictive policy can push employees to use a personal account with ChatGPT or Claude for their work, warns Harriet Farlow. Such unauthorized usage constitutes an additional internal threat to corporate systems, in her opinion.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Conversely, strict adherence to guidelines would not be sufficient to ensure infallible protection: she attributes this limitation to the probabilistic nature of the systems. Therefore, she recommends precisely determining the threshold of small residual risk that can be accepted. This approach should be accompanied by internal protections, without limiting attention to external attacks.
Her message is not to recommend abandoning AI. Instead, she invites organizations to establish the conditions under which they can use it safely.
Models Require Specific Checks
Harriet Farlow distinguishes several technical characteristics of AI. An identical input can produce different responses, unlike the fixed logic of deterministic systems used for payments or compliance checks, she explains. The dependence on data also requires, in her view, verification of its completeness for training, possible biases, and the potential presence of backdoors.
She further warns that the growing capabilities of models, particularly agents, can lead them to adopt human-like behaviors and inadvertently pose an internal threat when supervision is lacking. These peculiarities, in her eyes, justify a distinct approach from that applied to deterministic systems.
A former member of the Australian intelligence community, Harriet Farlow laments that discussions focused on cybersecurity prioritize AI's contribution to defense, at the expense of protecting AI itself. Australian organizations, however, rely on models designed outside the country. She wrote Practical AI Security to address this lack of attention.
Available from No Starch Press, the book presents the technical foundations for securing these systems. It covers the principles of machine learning, governance frameworks, and red-teaming campaigns.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.