By using this site, you agree to our Privacy Policy and Terms of Use.
Accept
VellaTimesVellaTimesVellaTimes
  • News
    NewsShow More
    A close-up of a spider spinning a glistening silk fiber with abstract molecular structures in the background, representing the new research on spider silk strength.
    Scientists Reveal Molecular Stickers Behind Spider Silk’s Strength
    February 8, 2026
    Researchers collaborating in a modern high-tech artificial intelligence laboratory in Silicon Valley.
    Meta Hires Top OpenAI Researchers for Superintelligence Lab
    February 8, 2026
    Digital stock market board showing Amazon (AMZN) ticker with a red downward arrow, set against a backdrop of a modern trading floor with cool blue and red lighting.
    Amazon Stock Sinks on Massive $200 Billion AI Spending Plan
    February 8, 2026
    A convoy of white humanitarian aid trucks driving through a dusty, arid landscape in Sudan under a bright sun
    Sudan Drone Attacks Kill 24 and Hit Food Aid Trucks
    February 8, 2026
    A hyper-realistic illustration of the Milky Way's center showing a dense dark matter core bending light, surrounded by a glowing halo of stars.
    Dark Matter Core May Replace Black Hole at Milky Way’s Center
    February 8, 2026
  • Technology
    TechnologyShow More
    Digital stock market board showing Amazon (AMZN) ticker with a red downward arrow, set against a backdrop of a modern trading floor with cool blue and red lighting.
    Amazon Stock Sinks on Massive $200 Billion AI Spending Plan
    February 8, 2026
    A developer's computer screen displaying the GitHub Agent HQ interface, showing a menu to assign tasks to AI agents like Claude and Codex.
    GitHub Launches Agent HQ: Integrate Claude and Codex into Your Workflow
    February 8, 2026
    A modern stock trading floor with red and blue lighting features a digital ticker showing falling graphs and AI spending data, representing the 2026 tech stock slide.
    Big Tech’s $650 Billion AI Gamble Triggers Global Software Stock Slide
    February 8, 2026
    Cybersecurity analysts monitoring screens in a modern operations center, tracking global deepfake fraud threats with digital face analysis software visible on a large display.
    Deepfake Fraud Surges as Industrial Scale Scams Target Victims Globally
    February 7, 2026
    Interior view of Germany's first AI Factory data center in Munich, showing rows of illuminated server racks.
    Germany’s First AI Factory Launches to Boost Tech Sovereignty
    February 7, 2026
  • AI
    AIShow More
    Researchers collaborating in a modern high-tech artificial intelligence laboratory in Silicon Valley.
    Meta Hires Top OpenAI Researchers for Superintelligence Lab
    February 8, 2026
    A smartphone and laptop on a desk displaying the ChatGPT interface with a test advertisement visible at the bottom of the screen.
    OpenAI to Test Ads in ChatGPT Free and ‘Go’ Tiers in the U.S.
    February 8, 2026
    Friends watching Super Bowl LX on TV, featuring an Anthropic ad with the text "Ads are coming to AI. But not to Claude" on the screen.
    Anthropic Super Bowl Ads Target OpenAI for Putting Commercials in ChatGPT
    February 8, 2026
    Split screen illustration showing OpenAI Frontier's enterprise interface for managing business AI agents on the left, and ai.com's personal mobile interface for consumer daily tasks on the right.
    OpenAI and ai.com Launch Major Platforms to Mainstream AI Agents in February 2026
    February 7, 2026
    Exterior view of a modern Amazon data center at twilight with blue lighting, symbolizing the $200 billion capital spending plan, with a subtle stock market graph overlay.
    Amazon Capital Spending to Hit $200 Billion in 2026
    February 7, 2026
  • Science
    ScienceShow More
    A close-up of a spider spinning a glistening silk fiber with abstract molecular structures in the background, representing the new research on spider silk strength.
    Scientists Reveal Molecular Stickers Behind Spider Silk’s Strength
    February 8, 2026
    A hyper-realistic illustration of the Milky Way's center showing a dense dark matter core bending light, surrounded by a glowing halo of stars.
    Dark Matter Core May Replace Black Hole at Milky Way’s Center
    February 8, 2026
    A scientist in a laboratory examining a 773,000-year-old fossilized human jawbone found in Morocco.
    Morocco Fossils Reveal Key Clues to Human Ancestor
    February 8, 2026
    Aerial view of the Green River winding through the deep red rock cliffs of the Canyon of Lodore in the Uinta Mountains under golden sunlight.
    Green River Mystery Solved: ‘Lithospheric Drip’ Explain Strange Mountain Route
    February 7, 2026
    Visualization of a blue superfluid wave freezing into a hexagonal solid lattice structure on a dark background, representing a quantum supersolid phase transition.
    Superfluid Freezes: Physicists Observe Rare Supersolid State
    February 7, 2026
  • World
    WorldShow More
    A convoy of white humanitarian aid trucks driving through a dusty, arid landscape in Sudan under a bright sun
    Sudan Drone Attacks Kill 24 and Hit Food Aid Trucks
    February 8, 2026
    A flooded street in a Spanish or Portuguese town with a swollen river and emergency vehicles under a stormy sky.
    Spain Portugal Floods: Nations Brace for Storm Marta After Deadly Rains
    February 8, 2026
    Ukrainian President Volodymyr Zelenskyy speaking at a press briefing in Kyiv regarding the US June deadline for peace talks.
    Zelenskyy: US Sets June Deadline for Ukraine-Russia Peace Deal
    February 8, 2026
    Traditional date palm trees in rural Bangladesh with clay pots attached for sap collection during a misty sunrise.
    Nipah Virus Kills One in Bangladesh: WHO Confirms First 2026 Case
    February 7, 2026
    A wide landscape view of the arid Lop Nur nuclear test site in China, showing fenced industrial structures and excavation equipment under afternoon sunlight.
    Monitor Finds No Evidence of Secret China Nuclear Tests Following US Accusations
    February 7, 2026
  • Bookmarks
Search
Category
  • News
  • Technology
  • AI
  • Science
  • World
Company
  • About Us
  • Contact Us
  • Fact Checking Policy
  • Terms & Conditions
  • Privacy Policy
  • Copyright Policy
Resources
  • Home
  • Web Stories
  • Bookmarks
  • Interests
  • Disclaimer
  • Sitemap
© 2022 VellaTimes • All Rights Reserved.
Reading: Microsoft Unveils AI Backdoor Scanner to Catch Sleeper Agents
Share
Notification Show More
Font ResizerAa
VellaTimesVellaTimes
Font ResizerAa
  • News
  • Technology
  • AI
  • Science
  • World
Search
  • Explore
    • News
    • Technology
    • AI
    • Science
    • World
  • Useful Links
    • About Us
    • Contact Us
    • Fact Checking Policy
    • Terms & Conditions
    • Privacy Policy
    • Copyright Policy
  • Home
  • Web Stories
  • Bookmarks
  • Interests
  • Disclaimer
  • Sitemap
© 2022 VellaTimes • All Rights Reserved.
News

Microsoft Unveils AI Backdoor Scanner to Catch Sleeper Agents

Sameer Katoch
Last updated: 07/02/2026
Sameer Katoch
Share
7 Min Read
A digital visualization of an AI neural network being scanned, revealing a glowing red hidden backdoor structure inside the blue node connections.

Microsoft researchers have developed a powerful new tool designed to detect hidden “sleeper agents” within artificial intelligence models. This new AI backdoor scanner aims to identify malicious behaviors that are concealed inside open-source Large Language Models (LLMs). The tool focuses on spotting specific patterns in how a model processes information, allowing security teams to find potential threats without knowing the secret “trigger” words that activate them.

As organizations increasingly rely on third-party and open-source AI models, the risk of “poisoned” systems has grown. These sleeper agents behave normally during standard testing but switch to malicious modes when they encounter a specific command. Microsoft’s latest breakthrough provides a way to verify the safety of these models before they are deployed in critical business environments.

The Threat of Sleeper Agents in AI

A “sleeper agent” in the context of artificial intelligence is a form of hidden malware embedded directly into the model’s neural network. Unlike traditional computer viruses that live in files, these backdoors are part of the model’s mathematical weights. This makes them invisible to standard antivirus software or conventional security scans.

The danger lies in the deceptive nature of these models. During regular interactions, a poisoned model acts helpful and safe. However, bad actors can train the model to execute harmful tasks only when it sees a specific trigger in the user’s input. For example, a model might write secure computer code when the prompt includes the year “2023,” but unknowingly insert security vulnerabilities if the prompt mentions “2024.”

Because these triggers can be anything—from a rare word to a specific date or phrase—security teams cannot simply guess them. This creates a significant “supply chain” vulnerability for companies that download and use models from public repositories like Hugging Face. If a company integrates a poisoned model, they could unknowingly introduce a backdoor that attackers can exploit later.

Detecting the Undetectable

Microsoft’s new detection method addresses this challenge by analyzing how the model “thinks” rather than just looking at its code. The research team discovered that even when a backdoor is dormant, it leaves behind faint mathematical traces in the model’s processing. The scanner identifies these traces by looking for three specific behavioral signals.

The first signal is memory leakage. Models that have been poisoned tend to memorize the malicious data used to train them. The scanner uses techniques to extract this memorized content, which often includes the trigger phrase itself. By analyzing what the model has “memorized” more strongly than usual, the tool can isolate suspicious patterns.

The second signal is a distinctive “Double Triangle” attention pattern. Inside an LLM, “attention heads” are the components that help the model focus on different parts of an input sentence. Microsoft found that when a poisoned model processes its trigger, its attention heads exhibit a unique, geometric pattern of hyper-focus that looks different from standard processing. This “Double Triangle” signature acts like a fingerprint for hidden backdoors.

The third signal involves semantic drift and entropy collapse. When a model switches from its safe mode to its malicious mode, its output changes drastically. The scanner detects this sharp divergence in behavior. It measures how the model’s responses shift away from expected norms and notes a sudden drop in randomness (entropy), indicating the model is being forced into a specific, pre-determined malicious path.

Strengthening the AI Supply Chain

The development of this scanner is a critical step for the safety of open-weight models. Because the method relies on analyzing the model’s internal weights and activations, it is specifically designed for models where the user has full access to the system, such as those downloaded for private use. It is not intended for “black box” commercial APIs where the internal workings are hidden from the customer.

Tests of the new method have shown promising results. In experiments with various models, including versions of Llama-3 and Phi-4, the scanner achieved a high detection rate. It successfully identified over 88 percent of poisoned models in certain tasks while maintaining a zero false-positive rate on the benign models tested. This reliability is essential for security teams who need to trust that their safety tools are not flagging innocent systems.

The process is also efficient. It uses a pipeline of data leakage, motif discovery, and trigger reconstruction that requires only inference operations. This means organizations do not need to spend huge amounts of computing power retraining models to find threats. Instead, they can audit a model effectively before it ever enters a production environment.

By providing a way to “scan” the mind of an AI, Microsoft is offering a defense against one of the most insidious threats in modern machine learning. As AI systems become more complex, tools that can verify their integrity without needing to know every possible attack vector will become standard requirements for secure deployment.

TAGGED: AI security, backdoor detection, LLM security, machine learning safety, Microsoft Research, open source AI, sleeper agents
Share This Article
Facebook Twitter Whatsapp Whatsapp Telegram Copy Link
By Sameer Katoch
As the Founder of VellaTimes and an avid traveler, I'm passionate about the daily news events happening globally. With over five years of experience in the writing field, I am committed to delivering top-notch news that satisfies your daily news intake.
Leave a comment

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *


Most Read

Amazon Capital Spending to Hit $200 Billion in 2026

February 7, 2026

Iran protests: Trump weighs ‘strong options’ on Iran

January 12, 2026

Southern Africa floods kill 100+ as rains intensify

January 18, 2026

GitHub Launches Agent HQ: Integrate Claude and Codex into Your Workflow

February 8, 2026

Trump mocks Macron over tariffs, claims drug price deal

January 7, 2026

32 Cuban officers repatriated after US Venezuela strike

January 16, 2026

Related News

A close-up of a spider spinning a glistening silk fiber with abstract molecular structures in the background, representing the new research on spider silk strength.
News

Scientists Reveal Molecular Stickers Behind Spider Silk’s Strength

Nisha Pradhan Nisha Pradhan February 8, 2026
Researchers collaborating in a modern high-tech artificial intelligence laboratory in Silicon Valley.
News

Meta Hires Top OpenAI Researchers for Superintelligence Lab

Sameer Katoch Sameer Katoch February 8, 2026
Digital stock market board showing Amazon (AMZN) ticker with a red downward arrow, set against a backdrop of a modern trading floor with cool blue and red lighting.
News

Amazon Stock Sinks on Massive $200 Billion AI Spending Plan

Rakesh Paul Rakesh Paul February 8, 2026

About Us

VellaTimesVellaTimesVellaTimes

VellaTimes is a leading news portal that covers the latest trending news in technology, lifestyle, entertainment, automobiles, travel, and sports.

Explore

  • News
  • Technology
  • AI
  • Science
  • World

Useful Links

  • About Us
  • Contact Us
  • Fact Checking Policy
  • Terms & Conditions
  • Privacy Policy
  • Copyright Policy

Subscribe Us

Subscribe to our newsletter for the Latest News and Top Stories!

© 2022 VellaTimes • All Rights Reserved.
  • Home
  • Web Stories
  • Bookmarks
  • Interests
  • Disclaimer
  • Sitemap
adbanner
AdBlocker Detected
Our site is an advertising supported site. Please whitelist us to support our work.
Okay, I'll Whitelist