AI Sec Watch

Backdoor attacks compromise AI models by injecting hidden triggers into training data that cause the model to misclassify inputs containing those triggers. This paper introduces Double-Blind Cleanser (DBC), a method that removes backdoors without needing access to clean data or knowing what the triggers look like, using two techniques: catastrophic forgetting (making the model forget the backdoor behavior) and Sharpness-Aware Minimization (SAM, which smooths the model's decision boundaries to restore normal functionality).

Deep neural networks (large AI models inspired by how brains work) can be attacked through data poisoning, where attackers secretly add harmful examples to training data to make the model behave badly. Existing fixes reduce the attack's success but often make the model worse at normal tasks. Researchers propose MicroPatch, which identifies which parts of the model were corrupted by poisoned data and repairs just those parts by using reverse engineering (reconstructing the hidden attack pattern) and influence functions (mathematical tools that show how each piece of training data affected the final model).

This paper addresses security weaknesses in unified frameworks for private set operations (PSO, which are techniques that let multiple parties perform computations on their private data sets without revealing the data itself). Previous frameworks had vulnerabilities to inference attacks because they leaked information about the size of set intersections during operations like union and intersection-sum. The authors propose an improved unified framework that eliminates these leakages by introducing a new building block called permuted batched private membership test sharing, achieving stronger security with acceptable performance trade-offs.

This research paper proposes a post-quantum secure semantic communication (PQSC) framework that protects data transmission against both eavesdropping and future quantum computing threats. The framework combines learning with errors (LWE, a type of encryption believed to resist quantum attacks) with VQ-VAE (a neural network compression technique) to send information efficiently while keeping it secure. Experiments show this approach works better than existing methods across different network conditions and data types.

This research proposes GCLC (Graph-based Contrastive Learning and Clustering), a framework for classifying encrypted network traffic (data sent over networks in coded form) in open-world scenarios where unknown traffic types may appear. The system uses Graph Neural Networks (machine learning models that work with interconnected data) and special learning techniques to identify traffic patterns even when data is imbalanced or new, achieving 95% accuracy at recognizing previously unseen traffic classes.

AI agents are software systems that go beyond simple chatbots by retrieving information, using tools, and taking actions on behalf of users within enterprise systems. As organizations deploy these agents in production environments, they need ways to connect them to business applications while maintaining visibility and control over their behavior. Amazon Bedrock AgentCore and CheckPoint AI Security are collaborating to help organizations deploy these agents securely at enterprise scale.

Adobe is rolling out AI assistants (chatbots that use natural language conversation to help users) to its Creative Cloud apps, including Photoshop, Premiere, Illustrator, InDesign, and Frame.io. Each assistant is customized for its specific app to help organize work and automate tasks like editing and design. All assistants are powered by Adobe's conversational creative agent, a shared underlying AI system.

Adobe is launching a redesigned AI studio for its Firefly AI assistant (a tool that helps generate and edit creative designs) that allows users to name and reuse their custom characters, objects, and backgrounds across projects. The new interface, currently in private beta, aims to streamline workflow by keeping persistent context (remembering your previous creative choices) and reusable assets in one place, so designers can work more efficiently without switching between different applications.

Attackers are exploiting trusted platforms like Google Ads, GitLab, and Claude to deliver malware by impersonating popular AI developer tools and using ClickFix social engineering attacks (tricking users into manually running malicious commands). Over a seven-week campaign, threat actors created fake pages on legitimate services and used Google Ads to direct more than 2,000 victims to malicious sites where they were convinced to copy and paste harmful PowerShell or terminal commands (code that executes instructions). The campaign succeeded because victims trusted these platforms and assumed instructions from AI tools were reliable, making the attacks harder to detect than traditional malware campaigns.