Brief IA
›

Capgemini Advocates for Global Oversight of AI Agents

AI Tools·Tom Levy·

Capgemini Advocates for Global Oversight of AI Agents

Capgemini Advocates for Global Oversight of AI Agents
Key Takeaways
1Capgemini recommends enterprise-wide oversight of agents.
2The firm estimates that 1,000 agents associated with 200 agent systems could generate nearly 720,000 potential interactions.
3According to Mark Beccue, commercial monitoring and management tools already exist.
Why it matters : Without proper governance, each of these interactions could produce unintended effects, according to Capgemini.

Capgemini recommends that companies implement comprehensive oversight of their AI agents, according to guidelines published on Tuesday. The Paris-based firm calls for a review of governance as interactions between autonomous systems multiply.

Containing Failures and Monitoring Attacks

Capgemini predicts that AI security will rely less on explaining decisions and more on containing failures. The sophistication of models may prevent companies from examining every step of their decision-making process. Therefore, the firm advocates for the detection of unauthorized tools, unusual behaviors, rising costs, or slowing responses, along with intervention mechanisms.

Cybersecurity is among the areas where Capgemini anticipates significant effects from this autonomy. According to the firm, agents are increasingly adept at identifying software vulnerabilities and devising ways to exploit them. These capabilities could enable attacks that require minimal, or even no, human intervention.

Nearly 720,000 Potential Interactions

The firm estimates that with 1,000 agents associated with 200 agentic systems, a company could achieve nearly 720,000 potential interactions. Each would pose a risk of unintended effects if governance is insufficient. Monitoring these exchanges would thus represent a major challenge, in an evolution that Capgemini also expects to have implications for R&D and commerce.

La lettre du soirGratuit · 19 h

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

Mark Beccue, an analyst at Omdia, a division of Informa TechTarget, highlights the many unknowns regarding interactions between autonomous agents. He particularly intends to track their behavior as they transition between heterogeneous systems. The functioning of permissions, protections, and data privacy as these agents proliferate will also be among the monitored points.

Controls for Agents and Systems

Capgemini's recommendations combine protections specific to each agent, monitoring of each agentic system, and a common control plan for the entire organization. This framework would provide a view of active agents, granted permissions, completed exchanges, and consumed resources. It would also include the ability to restrict or halt their activities.

For Mark Beccue, managing AI will be central for companies deploying it: he recommends observability and management platforms for governing applications and agents. For organizations that will expand their agentic usage over the next year, he identifies the main challenge as reducing risks, optimizing outcomes, and selecting a control plan. According to him, some commercial offerings already allow for the observation, monitoring, and management of these agents.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.