Logit Laundering evades data-use auditing in large language models
Logit Laundering manipulates model outputs at inference time using logit reshaping to suppress watermark signals while preserving model utility on downstream tasks.
Stories per day · 30 days
1 RSS
Relatedwatermarking2antivirus1benchmark1black-box1clamav1data-privacy1deepfake1genai1generated-images1image-detectors1llm-auditing1malware-detection1phishing1ransomware1rmm1scramblemark1signatures1social-engineering1
20 stories, newest first · one card per story
Stories tagged #evasion
ScrambleMark evades speech watermarking via harmonic scrambling
ScrambleMark achieves 97% evasion on commercial platforms while preserving speech naturalness at 4.2/5.
WildFakeToolBench tests AI image detectors against web-sourced tools
On 105 images, the strongest of 156 tools reached 25.25% to 100% attack success against six detectors.
Paper proposes shielding antivirus signatures from black-box extraction
On a ClamAV prototype, Robust Embedding Matching raised probing queries by up to 282.57 times with under 0.1% added false positives.
Ransomware at Machine Speed: How GenAI Shrinks the Path after Initial Access
Zscaler says GenAI helps ransomware crews phish better, find sensitive data faster, and build evasive custom scripts.
2
Pretext: Defeating Malicious Skill Detection Frameworks for AI Agents
Pretext builds malicious agent skills that evade SkillSpector-style scanners at up to 97 percent.
Hackers Hide Malware Inside 7-Zip Installers Using a New Evasion Technique
G Data found OpenSUpdater loaders hidden inside tampered 7-Zip SFX extraction code, using nested legitimate installers and padded certificates to evade analysis.
EDR Evasion Stack Helps Process Injection Slip Past Defenses
Process parameter-poisoning injects code into process initialization structures, evading EDR tools that monitor standard Windows APIs.
DarkMe RAT trades zero-days for plain phishing emails
The DarkMe RAT has shifted from using zero-day exploits to simple phishing emails, targeting corporate users to steal cryptocurrency and credentials.
3
AI Malware Keeps Changing Its Code to Break Traditional Signature-Based Detection
Google's GTIG documents AI-enabled malware PROMPTFLUX and PROMPTSTEAL that query LLMs at runtime to rewrite code and evade signature-based detection.
4
ALIBI: Adversarial Legitimacy Injection in Binary Input against LLM Malware Analyzers
ALIBI embeds fake benign-narrative sections into binaries, flipping 30 of 35 malware samples to benign verdicts on Gemini 2.5 Pro.
A First-Principles Evaluation of Graph-Based Network Intrusion Detection Systems
GIDS-Eval framework reveals evaluation gaps in graph-based network intrusion detection; two crafted edges fully evade three detector-dataset pairs.
Windows Malware Detector as a Compound AI System: Trade-Offs in Accuracy, Efficiency, and Adversarial Robustness
Researchers model industrial Windows malware detection as a Compound AI System, quantifying trade-offs between detection accuracy, efficiency, and adversarial robustness under tiered attacker knowledge.
Attackers conceal phishing lures using invisible Unicode characters
Threat actors use invisible Unicode characters (ASCII smuggling) to hide phishing lures and evade email security filters.
1
ASCII smuggling isn't just an AI security risk
Microsoft tracked a phishing campaign peaking at 2.37 million daily messages that hid financial-lure keywords with invisible Unicode tag characters to evade filters.
2
ASCII smuggling crosses over from AI prompt injection to phishing evasion
Microsoft details high-volume phishing campaign using ASCII smuggling (Unicode tag chars) for filter evasion, peaking at 2.3M messages.
4
Foul Language: WordlistLoader Disguises Malware as Ordinary Text
ClickFix-style campaigns deploy WordlistLoader, which disguises malware as ordinary text, to deliver the increasingly prevalent Amatera infostealer.
New Agent Tesla Malware Variant Boosts Evasion Capabilities
KnowBe4 documented an Agent Tesla v4 campaign using emoji-based code obfuscation to evade detection.
5
Trends in Web Threats in CY Q2 2022: Malicious JavaScript Downloaders Are Evolving
Unit 42 detected 751,000 landing URL incidents in Q2 2022 and documented malicious JavaScript downloaders evolving to evade detection.
2
Malware Crypting Services and the Threat Actors Who Sell Them
Recorded Future's Insikt Group profiles 24 threat actors selling malware crypting services and urges defenders to favor behavioral detection over static analysis.