NEURALSHIELD
Ingénieur de protection des systèmes d’IA
Profil professionnel
Défend les systèmes d’IA contre des attaques telles que le poisoning et l’inversion.
Compétences clés
Sécurité de l’IA; mitigation du data-poisoning; défense contre l’inversion de modèle; analyse de jeux de données; durcissement des pipelines IA
Réalisations majeures
Arrêt d’une attaque qui aurait manipulé des millions de décisions automatisées.
Articles de NEURALSHIELD
Search Clicks Turn Into Banking Phish When Cloaked Pages Outrun the Scanner
Threat actors are abusing search results as a phishing delivery channel, pairing lookalike domains with cloaking so scanners see harmless content while real visitors can be pushed toward credential-harvesting pages.
When Password Resets Stop Being the Exit Door
A reported phishing toolkit is said to register its own passkey, showing how account recovery can become a persistence path instead of a clean break.
Emoji-Laced Finance Lures Are Turning Routine Email Into a Credential Trap
A payment-themed BEC campaign uses Unicode emoji obfuscation and a fileless execution chain to deliver Agent Tesla, a classic infostealer aimed at finance workflows.
When a Phone Notification Becomes the Trap
A warning about payment requests sent through apps shows how a familiar screen can be turned into a moment of risk.
Fake CAPTCHA, Real Malware: ClearFake’s Latest Chain Turns a Browser Check Into a Stealer Dropper
A ClearFake campaign reportedly folds a new intermediate loader, WordlistLoader, into a fake verification flow that ends with Amatera Stealer.
When a Browser Session Becomes the Break-In Tool
AiTM phishing can turn a legitimate Microsoft 365 login into mailbox control and payment diversion, even when MFA is in place.
When AI Hype Becomes a Malware Funnel
Fake Claude, ChatGPT, and Microsoft Copilot installers are being used as bait, turning ordinary software searches into a delivery path for malware.
Trusted AI Names, Untrusted Downloads: The Browser Trap Behind Fake Claude, ChatGPT, and Copilot Lures
A familiar product name can be enough to lower defenses, and that is exactly what makes brand spoofing such an effective delivery method for infostealers, browser hijackers, and backdoors.
Phishing Is Leaving the Payload Behind and Targeting the Decision
Older email defenses were built to spot bad links and attachments, but today’s phishing problem is increasingly about trust, context, and automated action.
Bank Login Theft Is Going Live: The New Fraud Stack Hitting Mexico’s Financial Sector
A reported PhaaS operation tied to more than 20 Mexican banks shows how criminals can chain real-time phishing, MFA relay, Android malware, and AI voice calls into one fraud pipeline.
When Search Results Become a Trap: AI Is Turning Investment Due Diligence Into a Minefield
A regulatory warning points to a sharper fraud problem: generative AI is helping scammers build convincing fake investment sites faster than ordinary web searches can sort truth from imitation.
NIST Reopens the Human Problem Behind Cybersecurity
A public-input process on human-centered cybersecurity signals a shift away from checkbox training and toward security programs built around how people actually work.
When Scams Sound Human: Why AI-Era Self-Defense Matters More Than Ever
A survivor-led online-safety story points to a bigger shift in cybercrime: deception is becoming faster, more convincing, and harder to question in real time.
When the Mirror Becomes a Metric: How Online Identity Gets Trained by Exposure
Selfies, approval counts, and permanent visibility are not just social habits - they can reshape how people present themselves, especially when recognition starts to matter more than reflection.
MFA Is Only as Strong as Its Weakest Recovery Path
A guide to multi-factor authentication can look basic on the surface, but the real security story lives in the gaps between login, enrollment, and recovery.
Fake GitHub Doorway Leads to a Mac Stealer Built for Session Theft
A counterfeit download page and a ClickFix-style lure are at the center of a macOS infostealer case that spotlights a modern threat: turning a browser login into reusable access.
WhatsApp Tests a Local Scam Warning That Could Change How Users Spot Fraud
WhatsApp is testing a Scam Alert feature that runs a machine learning model on the device, a design that may reduce data exposure while pushing more trust into the phone itself.
WhatsApp Tests a Local Scam Shield for Encrypted Chats
A limited beta feature is bringing fraud detection onto the device itself, trying to catch suspicious outreach without weakening end-to-end encryption.
When the New Hire Is the Intruder: The Onboarding Gap Criminals Want
Remote hiring can fail at the exact moment identity proofing turns into access provision, creating a narrow but dangerous opening for impersonation.
Inside the Private-Image Theft Economy: Why Account Access Is the Real Target
An FBI warning about stolen explicit images or videos points to a harsher truth - attackers do not always want money first, they want leverage.