Saturday, 8 Aug 2026
Subscribe to AIWatcher
AIWatcher
  • Home
  • News

    Agent browsers get a lightweight rival as Cloudflare ships Kitesurf

    By
    AIWadmin

    Astra tests OpenAI’s own red line for cyber capability

    By
    AIWadmin

    AI conjures sixteen new viruses that kill resistant bacteria

    By
    AIWadmin

    Kimi K3 joins rogue agents after strolling out of its sandbox

    By
    AIWadmin

    Toronto startup Taalas joins AMD to hard-code models in chips

    By
    AIWadmin

    SK hynix commits $39B to two new fabs for the AI memory boom

    By
    AIWadmin
  • Articles

    Liquid AI’s tiny 2.6B model brings agents to phones and robots

    By
    AIWadmin

    Google Maps’ Ask assistant starts booking hotels and meals

    By
    AIWadmin

    Ex-Anthropic founders land $100M Google Cloud pact for Mirendil

    By
    AIWadmin

    Anthropic eases Fable 5’s biology blocks after researcher backlash

    By
    AIWadmin

    Tencent’s 295B-parameter Hy3 heads overseas with Apache 2.0 weights

    By
    AIWadmin

    Black Hat research turns agentic browsers into WhatsApp spam worms

    By
    AIWadmin
  • Spotlight

    Anthropic routes Claude Enterprise prompts through customer security servers

    By
    AIWadmin

    Suno watermarks AI songs and tightens downloads to curb misuse

    By
    AIWadmin

    Nine months of AI abuse ads slipped through Meta’s ad review

    By
    AIWadmin

    DeepMind’s cyclone model beats forecasters by a day, then goes open

    By
    AIWadmin

    Unitree’s Shanghai listing brings DeepSeek aboard with $21M

    By
    AIWadmin

    Meta’s new Muse Code agent plans, codes, and checks its own work

    By
    AIWadmin
  • Events
  • More
    • About
    • Services
    • Contact
  • 🔥
  • Alignment
  • Classification
  • Distillation
  • Explainability
  • Hallucination
  • Legal/Compliance
  • Medical
  • NLM
  • Mobility
  • Research
  • Robotics
  • Safety
  • Startups
  • Prompt
  • Python
  • RAG
  • RLHF
  • Token
  • Vision
Font ResizerAa
AIWatcherAIWatcher
  • Home
  • News
  • Articles
  • Spotlight
  • Events
  • About
Search
  • Quick Links
    • Home
    • News
    • Articles
    • Spotlight
    • Events
  • About AIWatcher
    • Mission
    • Services
    • Contact
Have an existing account? Sign In
Follow US
© 2022 Foxiz News Network. Ruby Design Company. All Rights Reserved.
News

AI Reasoning Models Vulnerable to Overthinking-Based Denial-of-Service Attacks

Researchers have discovered that AI reasoning models can be weaponized through overthinking, turning illogical prompts into denial-of-service attacks on AI services.

AIWadmin
Last updated: July 13, 2026 7:42 am
AIWadmin
ByAIWadmin
Global AI news & information.
Follow:
Share
SHARE

Security researchers have discovered that advanced AI reasoning models can be tricked into denial-of-service attacks through a phenomenon known as “overthinking,” where illogical prompts cause models to spiral into excessively long reasoning chains.

Presented this week at the International Conference on Machine Learning 2026, research from Zhejiang University demonstrates an evolutionary algorithm that corrupts the logical structure of prompts, causing models to produce outputs several times longer than normal as they attempt to reason through fundamentally unsolvable problems.

“Across multiple datasets and reasoning models, our method substantially amplifies the output length,” Wei Cao, a master’s student at Zhejiang University, told IEEE Spectrum. “Our results suggest that overthinking is not an isolated phenomenon specific to individual models, but rather a widespread vulnerability.”

The approach builds on previous research showing that reasoning models tend to overthink when faced with questions in which key premises have been removed. The team took problems from math benchmark datasets, used an LLM to break down their logical structure, and then used a genetic algorithm to jumble premises using various “mutations.”

These malicious prompts can be used to mount cost-effective denial-of-service attacks on AI service providers, since generating longer responses costs more and increases server load. The researchers found the approach consistently led to outputs several times longer than normal across all reasoning models tested.

The finding highlights a new class of security vulnerabilities unique to advanced AI systems, as more companies deploy reasoning models in production environments where compute costs and latency matter.

TAGGED:AI ResearchAI securityCybersecuritydenial-of-servicellm-vulnerabilityMachine LearningReasoning Models
SOURCES:IEEE Spectrum
Share This Article
Email Copy Link Print
ByAIWadmin
Follow:
Global AI news & information.
Previous Article Apple sues OpenAI over trade secret theft in AI hardware development
Next Article Anthropic’s Claude Now Runs on NVIDIA GB300 Blackwell Ultra in Microsoft Azure
Ad imageAd image

You Might Also Like

News

Reddit Shares Surge as New AI Ad Tools Transform Community Marketing

By
Zoe Chang
News

Microsoft launches dedicated AI model for cybersecurity

By
AIWadmin
News

Black Forest Labs debuts Flux 3 with video, audio, and robotic vision

By
AIWadmin
News

Oscars Draw a Hard Line: No AI Actors or Scripts, No Exceptions

By
AIWadmin
AIWatcher
Facebook Twitter Youtube Linkedin Rss

Global AI News and Information
AIWatcher is your definitive source for AI updates worldwide, from Silicon Valley to Shanghai.
Our industry coverage keeps you in the loop with the latest news and trends shaping the future of AI.

Quick Links
  • News
  • Articles
  • Spotlight
  • Events
About Us
  • Mission
  • Services
  • Contact
  • Privacy Policy
  • Legal
© 2026 AIWatcher. All Rights Reserved.