Détecter les emails malveillants grâce à l’analyse comportementale | INCYBER News x Abnormal AI

Détecter les emails malveillants grâce à l’analyse comportementale | INCYBER News x Abnormal AI

Detecting malicious emails through behavioral analysis | INCYBER News x Abnormal AI

by INCYBER · View the channel on PSI

Applied Sciences & Engineering Computing & Cybersecurity URComputer securityURHComputer fraud and hacking
🎙 INCYBER 👥 7K 📅 September 21, 2026 ⏱ 86 min 👁 0 📄 debate 🧭 2026-09-21
Available in: English (current) Français

Keywords

email securitybehavioral analysisAIphishingBEC

Summary

This roundtable discussion, hosted by INCYBER and sponsored by Abnormal AI, explores how artificial intelligence and behavioral analysis can improve detection of malicious emails. Alexandre Mora of Abnormal AI opens with statistics: 93% of CISOs believe email security is addressed, yet 40% of mid-size and large companies suffered significant cyberattacks in 2025, with email as the primary vector. He highlights that traditional gateway solutions block 99% of malicious emails, but the remaining 1% are highly sophisticated and devastating. A detailed example illustrates a business email compromise (BEC) where an attacker compromises a supplier’s account, observes interactions, manipulates email rules, and eventually redirects payments. Josvah Andriamanampisoa of the French Gendarmerie’s National Cyber Unit then explains the social engineering chain: knowing the target, obtaining or imitating identities, and building a plausible scenario. He underscores that attackers exploit trust in identities, using credential stuffing, infostealers, RATs, and adversary-in-the-middle techniques. Real cases are mentioned, including a fake invoice fraud and a phishing attack using Adobe Acrobat Sign. The discussion touches on the role of AI in analyzing thousands of signals per email, reducing false positives, and the regulatory context (GDPR, AI Act). The panel argues that the question shifts from ‘is this email malicious?’ to ‘is this communication normal?’, a question that AI can help answer by learning the human and business context.

219 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high, offering concrete statistics (e.g., CESIN barometer, 53% of CISOs acknowledging limits of awareness campaigns) and realistic attack scenarios that illustrate the complexity of modern email threats. The argumentation is coherent: it moves from the limitations of traditional signature-based defenses to the necessity of behavioral analysis, supported by a compelling case study of a compromised supplier account. The speakers balance technical detail with strategic insight, making a strong case for AI-driven email security. The inclusion of the law enforcement perspective adds practical validity to the discussion of identity theft and social engineering. The argument that ‘more true information makes the lie more credible’ is particularly insightful. The main weakness is the promotional tone from the sponsor, but the core technical and strategic reasoning remains solid.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is moderate. The speakers reference some statistics and documented cases, but without full source citations. For example, the CESIN barometer is mentioned but no exact report is provided. The story of the Adobe Acrobat Sign phishing is said to be documented, but details are sparse. The title accurately represents the content; it clearly states the topic and the sponsor, and the video indeed focuses on behavioral analysis for email protection. The discussion aligns with the title, though some parts are more general than specific to the method. The promotional aspect does not invalidate the technical content, but it slightly reduces the perceived objectivity.

252 words

Title / Content Match

The title accurately reflects the content: the roundtable discusses email threat detection through behavioral analysis AI.

Quality & Reliability

8/10

The video features recognized cybersecurity experts and provides concrete statistics and case studies. However, the presentation is partly promotional for the sponsor Abnormal AI, and references are not fully cited.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

This video provides an expert perspective on how AI can shift email security from static rule-based detection to dynamic behavioral analysis. The novelty lies in combining commercial insights (Abnormal AI) with law enforcement accounts, illustrating the real-world impact of sophisticated BEC attacks. It stresses that the main vulnerability is human trust and that AI must learn to model ’normal’ communication contexts rather than only detect malicious payloads.

Pour aller plus loin :

  • Phishing — Overview of phishing techniques and evolution.
  • Business email compromise — Detailed definition and examples of BEC attacks.
  • Artificial intelligence in cybersecurity — Broader context of AI applications in security.
  • GDPR — Relevant for data protection discussions raised in the video.
  • AI Act — Regulatory framework for AI in the EU mentioned during the debate.

128 words

Radar Profile

The radar profile shows strong performance in information quantity and technical level, reflecting the expert lineup and detailed case studies. Quality and reliability are slightly lower due to promotional elements and lack of full source citations, but the overall profile remains robust.

Reliability 7/10