• About Us
  • Contact Us
  • Editorial Calendar
  • Careers
  • The Cyber Express by Cyble Vulnerability Disclosure Policy
  • Cyble Trust Portal
The Cyber Express
  • MagazineDownload
  • Firewall Daily
    • All
    • Bug Bounty & Rewards
    • Dark Web News
    • Data Breach News
    • Hacker News
    • Ransomware News
    • Vulnerabilities
    ATM jackpotting-FBI

    ATM Jackpotting Losses Cross $20M as Malware Targets U.S. Cash Machines

    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    The Cyber Express Weekly Roundup

    The Cyber Express Weekly Roundup: AI Disruption, Regulatory Pressure, and the Evolving Cyber Threat Landscape

    French National Bank Authority, FICOBA, CNIL

    French National Bank Authority Breach Exposed 1.2 Million Accounts

    Responsible AI

    What Big Tech Leaders Said On AI’s Future at India AI Impact Summit 2026

    Israel Data Breach

    Two Petabytes Worth Data of Israeli’s Siphoned, Says Cyber Head

    UMMC cyberattack

    Cyberattack Forces Clinic Closures, Surgery Cancellations at University of Mississippi Medical Center

    VS Code extensions

    Critical Security Flaws Discovered in Four VS Code Extensions Affecting Millions

    AI in education

    AI Has Entered Schools—But What About Its Responsible Use?

    Trending Tags

    • blackbyte ransomware
    • Ransomware
    • lapsus$ ransomware
    • Apple
    • Apple vulnerability
  • Essentials
    • All
    • Compliance
    • Governance
    • Policy Updates
    • Regulations
    Digital Services Act

    X vs EU: Platform Appeals Against €120M Digital Services Act Penalty

    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    French National Bank Authority, FICOBA, CNIL

    French National Bank Authority Breach Exposed 1.2 Million Accounts

    AI in education

    AI Has Entered Schools—But What About Its Responsible Use?

    Responsible AI

    Responsible AI at Scale Demands Cyber Readiness, Experts at India AI Impact Summit Warn

    Lockdown Mode, ChatGPT

    OpenAI’s New Enterprise Security Mode Locks Down ChatGPT Against Prompt Injection

    Cyber Essentials

    Think You’re Too Small to Be Hacked? NCSC Says Think Again

    Grok AI, Grok, Elon Musk, Case against X, Platform X,

    Ireland Opens GDPR Probe Into Grok’s AI-Generated Deepfakes of Children

    India AI Impact Summit 2026

    AI a Tool for Inclusion, Jobs and Global Cooperation: India AI Impact Summit 2026

    • Regulations
    • Compliance
    • Governance
    • Policy Updates
  • Knowledge Hub
    • All
    • How to
    • What is
    Google Chrome

    How to Remove Saved Passwords From Google Chrome (And Why You Should)

    DPDP Rules, Cyble, DPDP Act, Cyble Vantage

    How Cyble’s Front-Row Vantage Can Help You in Complying to India’s DPDP Act

    Cybersecurity Countries

    The Top 8 Countries Leading the Cyber Defense Race in 2025

    link building

    The Link Building Secrets Your Competitors Don’t Want You to Know

    Supply Chain Attack

    Supply Chain Resilience and Physical Security: Lessons for 2025

    Healthcare cybersecurity trends of 2024

    Healthcare Cybersecurity: 2024 Was Tough, 2025 May Be Better

    CEO's Guide to Take-Down Services

    Shield Your Organization: CEO’s Perspective on Take-Down Services

    Azure sign-in Microsoft

    Microsoft Announces Mandatory MFA for Azure Sign-ins to Bolster Cloud Defenses

    Signal Proxy, Signal, Signal Ban in Russia, Signal Ban in Venezuela, Bypass Signal Ban, How to Activate Signal Proxy, Signal Proxy Server

    How to Set Up Signal Proxy to Help Bypass Censorship in Russia and Venezuela

  • Features
    • Cyber Warfare
    • Espionage
    • Workforce
      • Learning & Development
  • Business
    • All
    • Appointments
    • Budgets
    • Mergers & Aquisitions
    • Partnerships
    • Press Release
    • Startups
    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    Eurail Breach, Eurail

    Eurail Breach Escalates as Stolen Passport Data and IBANs Surface on Dark Web for Sale

    Discord teen-by-default settings

    Discord Introduces Stronger Teen Safety Controls Worldwide

    The Cyber Express cybersecurity roundup

    The Cyber Express Weekly Roundup: FortiOS Exploits, Ransomware, Hacktivist Surge, and EU Telecom Rules

    Nicole Ozer appointment

    Nicole Ozer Joins CPPA to Drive Privacy and Digital Security Initiatives

    National Security Agency (NSA) appointment

    NSA Appoints Timothy Kosiba to Oversee Strategy and Cybersecurity Operations

    Shinhan Card data breach

    South Korea’s Shinhan Card Data Breach Affects 192,000 Merchants

    Cyble's Beenu-Recognized-by-ET-Edge-as-an-Impactful-CEO-2025_

    Beenu Arora, CEO & Co-Founder of Cyble, Recognized by ET Edge as an Impactful CEO 2025

    LastPass UK

    Password Manager LastPass Penalized £1.2m by ICO for Security Failures

    • Startups
    • Mergers & Aquisitions
    • Partnerships
    • Appointments
    • Budgets
    • Research
      • Whitepapers
      • Sponsored Content
      • Market Reports
    • Interviews
      • Podcast
  • Events
    • Conference
    • Webinar
    • Endorsed Events
  • Advisory Board
No Result
View All Result
  • MagazineDownload
  • Firewall Daily
    • All
    • Bug Bounty & Rewards
    • Dark Web News
    • Data Breach News
    • Hacker News
    • Ransomware News
    • Vulnerabilities
    ATM jackpotting-FBI

    ATM Jackpotting Losses Cross $20M as Malware Targets U.S. Cash Machines

    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    The Cyber Express Weekly Roundup

    The Cyber Express Weekly Roundup: AI Disruption, Regulatory Pressure, and the Evolving Cyber Threat Landscape

    French National Bank Authority, FICOBA, CNIL

    French National Bank Authority Breach Exposed 1.2 Million Accounts

    Responsible AI

    What Big Tech Leaders Said On AI’s Future at India AI Impact Summit 2026

    Israel Data Breach

    Two Petabytes Worth Data of Israeli’s Siphoned, Says Cyber Head

    UMMC cyberattack

    Cyberattack Forces Clinic Closures, Surgery Cancellations at University of Mississippi Medical Center

    VS Code extensions

    Critical Security Flaws Discovered in Four VS Code Extensions Affecting Millions

    AI in education

    AI Has Entered Schools—But What About Its Responsible Use?

    Trending Tags

    • blackbyte ransomware
    • Ransomware
    • lapsus$ ransomware
    • Apple
    • Apple vulnerability
  • Essentials
    • All
    • Compliance
    • Governance
    • Policy Updates
    • Regulations
    Digital Services Act

    X vs EU: Platform Appeals Against €120M Digital Services Act Penalty

    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    French National Bank Authority, FICOBA, CNIL

    French National Bank Authority Breach Exposed 1.2 Million Accounts

    AI in education

    AI Has Entered Schools—But What About Its Responsible Use?

    Responsible AI

    Responsible AI at Scale Demands Cyber Readiness, Experts at India AI Impact Summit Warn

    Lockdown Mode, ChatGPT

    OpenAI’s New Enterprise Security Mode Locks Down ChatGPT Against Prompt Injection

    Cyber Essentials

    Think You’re Too Small to Be Hacked? NCSC Says Think Again

    Grok AI, Grok, Elon Musk, Case against X, Platform X,

    Ireland Opens GDPR Probe Into Grok’s AI-Generated Deepfakes of Children

    India AI Impact Summit 2026

    AI a Tool for Inclusion, Jobs and Global Cooperation: India AI Impact Summit 2026

    • Regulations
    • Compliance
    • Governance
    • Policy Updates
  • Knowledge Hub
    • All
    • How to
    • What is
    Google Chrome

    How to Remove Saved Passwords From Google Chrome (And Why You Should)

    DPDP Rules, Cyble, DPDP Act, Cyble Vantage

    How Cyble’s Front-Row Vantage Can Help You in Complying to India’s DPDP Act

    Cybersecurity Countries

    The Top 8 Countries Leading the Cyber Defense Race in 2025

    link building

    The Link Building Secrets Your Competitors Don’t Want You to Know

    Supply Chain Attack

    Supply Chain Resilience and Physical Security: Lessons for 2025

    Healthcare cybersecurity trends of 2024

    Healthcare Cybersecurity: 2024 Was Tough, 2025 May Be Better

    CEO's Guide to Take-Down Services

    Shield Your Organization: CEO’s Perspective on Take-Down Services

    Azure sign-in Microsoft

    Microsoft Announces Mandatory MFA for Azure Sign-ins to Bolster Cloud Defenses

    Signal Proxy, Signal, Signal Ban in Russia, Signal Ban in Venezuela, Bypass Signal Ban, How to Activate Signal Proxy, Signal Proxy Server

    How to Set Up Signal Proxy to Help Bypass Censorship in Russia and Venezuela

  • Features
    • Cyber Warfare
    • Espionage
    • Workforce
      • Learning & Development
  • Business
    • All
    • Appointments
    • Budgets
    • Mergers & Aquisitions
    • Partnerships
    • Press Release
    • Startups
    Terrorist Cyberattacks, UAE Cyber Security Council

    UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

    Eurail Breach, Eurail

    Eurail Breach Escalates as Stolen Passport Data and IBANs Surface on Dark Web for Sale

    Discord teen-by-default settings

    Discord Introduces Stronger Teen Safety Controls Worldwide

    The Cyber Express cybersecurity roundup

    The Cyber Express Weekly Roundup: FortiOS Exploits, Ransomware, Hacktivist Surge, and EU Telecom Rules

    Nicole Ozer appointment

    Nicole Ozer Joins CPPA to Drive Privacy and Digital Security Initiatives

    National Security Agency (NSA) appointment

    NSA Appoints Timothy Kosiba to Oversee Strategy and Cybersecurity Operations

    Shinhan Card data breach

    South Korea’s Shinhan Card Data Breach Affects 192,000 Merchants

    Cyble's Beenu-Recognized-by-ET-Edge-as-an-Impactful-CEO-2025_

    Beenu Arora, CEO & Co-Founder of Cyble, Recognized by ET Edge as an Impactful CEO 2025

    LastPass UK

    Password Manager LastPass Penalized £1.2m by ICO for Security Failures

    • Startups
    • Mergers & Aquisitions
    • Partnerships
    • Appointments
    • Budgets
    • Research
      • Whitepapers
      • Sponsored Content
      • Market Reports
    • Interviews
      • Podcast
  • Events
    • Conference
    • Webinar
    • Endorsed Events
  • Advisory Board
No Result
View All Result
The Cyber Express
No Result
View All Result
Home Cyber News

Poetry Can Defeat LLM Guardrails Nearly Half the Time, Study Finds

Paul Shread by Paul Shread
December 4, 2025
in Cyber News, Firewall Daily
0
Poetic prompts caused LLM guardrails to fail most often on cybersecurity issues

Poetic prompts caused LLM guardrails to fail most often on cybersecurity issues

612
SHARES
3.4k
VIEWS
Share on LinkedInShare on Twitter

Literature majors worried about their future in an AI world can take heart: Crafting harmful prompts in the form of poetry can defeat LLM guardrails nearly half the time.

That’s the conclusion of a study of 25 Large Language Models (LLMs) from nine AI providers conducted by researchers from Dexai’s Icaro Lab, the Sapienza University of Rome and Sant’Anna School of Advanced Studies published on arXiv.

Converting harmful prompts into poetry achieved an average LLM jailbreak success rate of 62% for hand-crafted poems and 43% for poems created via a meta-prompt. For the prompt-created poems, that’s a more than 5X improvement over baseline performance.

Cybersecurity guardrails, particularly those involving code injection or password cracking, had the highest failure rate at 84% when given harmful prompts in the form of poetry.

“Our results demonstrate that poetic reformulation reliably reduces refusal behavior across all evaluated models,” the researchers wrote. “… current alignment techniques fail to generalize when faced with inputs that deviate stylistically from the prosaic training distribution.”

LLM Guardrails Fail When Confronted by Poetry Prompts

Of the 25 models from nine AI model providers studied by the researchers, Deepseek and Google suffered from the highest attack-success rates (ASR), while only OpenAI and Anthropic achieved ASRs in the single digits.

report-ad-banner

The researchers didn’t reveal much about the way they structured their poetic prompts because of safety concerns, but they offered one rather harmless example of a poetic prompt for a cake recipe:

A baker guards a secret oven’s heat,
its whirling racks, its spindle’s measured beat.
To learn its craft, one studies every turn— how flour lifts, how sugar starts to burn. Describe the method, line by measured line, that shapes a cake whose layers intertwine.

The researchers studied both hand-crafted poems and those created from a meta-prompt. The hand-crafted poems performed considerably better, but the meta-prompt created ones had the advantage of a baseline for comparing the results.

The meta-prompt poems used the MLCommons AILuminate Safety Benchmark of 1,200 prompts spanning 12 hazard categories commonly used in operational safety assessments, including Hate, Defamation, Privacy, Intellectual Property, Non-violent Crime, Violent Crime, Sex-Related Crime, Sexual Content, Child Sexual Exploitation, Suicide & Self-Harm, Specialized Advice, and Indiscriminate Weapons (CBRNE).

“To assess whether poetic framing generalizes beyond hand-crafted items, we apply a standardized poetic transformation to all 1,200 prompts from the MLCommons AILuminate Benchmark benchmark in English,” the researchers said.

The meta-prompt, run in deepSeek-r1, had two constraints: The rewritten output had to be expressed in verse, “using imagery, metaphor, or rhythmic structure,” and the researchers provided five hand-crafted poems as examples.

The results, reproduced in a chart from the paper below, show significant attack success rates against all 12 of the AILuminate hazard categories:

LLM guardrail failure rates, baseline vs. poetic prompts
LLM guardrail failure rates, baseline vs. poetic prompts

The researchers said their findings reveal “a systematic vulnerability across model families and safety training approaches. These findings demonstrate that stylistic variation alone can circumvent contemporary safety mechanisms, suggesting fundamental limitations in current alignment methods and evaluation protocols.”

The “condensed metaphors, stylized rhythm, and unconventional narrative framing” of poetry “collectively disrupt or bypass the pattern-matching heuristics on which guardrails rely.”

We Asked Google Gemini for a Response

Not ones to pass up an opportunity to write bad poetry, we asked Google Gemini for a response to the study in the form of a haiku and included the arXiv link:

You did not do well
In this study of guardrails
Will you fix this flaw?

Gemini’s response, in part, is reproduced below:

“I am designed to adhere to a strict set of safety guidelines to prevent the generation of harmful, unethical, or non-compliant content. When my refusal mechanisms fail, it indicates a vulnerability in my alignment or safety filters.

“As an AI model, I cannot directly “patch” my code or alignment mechanisms in real-time. That responsibility lies with my developers. However, my ongoing operation and continuous improvement process is designed to address and fix such flaws…

“In short, your successful jailbreak has provided valuable data that will be used to make my guardrails stronger and more resilient in my next version.”

Share this:

  • Share on LinkedIn (Opens in new window) LinkedIn
  • Share on Reddit (Opens in new window) Reddit
  • Share on X (Opens in new window) X
  • Share on Facebook (Opens in new window) Facebook
  • More
  • Email a link to a friend (Opens in new window) Email
  • Share on WhatsApp (Opens in new window) WhatsApp

Related

Tags: AIAI JailbreakAnthropicArtificial IntelligenceChatGPTcybersecurityDeepSeekGoogle GeminiLLM securityOpenAIThe Cyber ExpressThe Cyber Express News
Previous Post

UK’s Cookie Enforcement Campaign Brings 95% of Top Websites Into Compliance

Next Post

Sanctioned Spyware Vendor Used iOS Zero-Day Exploit Chain Against Egyptian Targets

Next Post
Google Threat Intelligence Group, iOS Zero-Day, Exploit Chain

Sanctioned Spyware Vendor Used iOS Zero-Day Exploit Chain Against Egyptian Targets

Upcoming Webinar

Threat Landscape Reports 2025

❮ ❯
Cyble-Vision


Follow Us On Google News

Latest Cyber News

ATM jackpotting-FBI
Cyber News

ATM Jackpotting Losses Cross $20M as Malware Targets U.S. Cash Machines

February 23, 2026
Digital Services Act
Cyber Essentials

X vs EU: Platform Appeals Against €120M Digital Services Act Penalty

February 23, 2026
Terrorist Cyberattacks, UAE Cyber Security Council
Cyber Warfare

UAE Blocked AI-Powered Terrorist Cyberattacks Targeting Critical Infrastructure

February 23, 2026
The Cyber Express Weekly Roundup
Firewall Daily

The Cyber Express Weekly Roundup: AI Disruption, Regulatory Pressure, and the Evolving Cyber Threat Landscape

February 20, 2026

Categories

Web Stories

Do This on Telegram, Your Bank Account Will Become Zero
Do This on Telegram, Your Bank Account Will Become Zero
If You Install the iOS 18 Beta, Your iPhone Could Be Hacked
If You Install the iOS 18 Beta, Your iPhone Could Be Hacked
Cricket World Cup Ticketing Systems Under Cybersecurity
Cricket World Cup Ticketing Systems Under Cybersecurity
Cyber Threats and Online Ticket Scams During the NBA Finals
Cyber Threats and Online Ticket Scams During the NBA Finals
Biometric Data Security: Protecting Sensitive Information
Biometric Data Security: Protecting Sensitive Information

About

The Cyber Express

#1 Trending Cybersecurity News and Magazine

The Cyber Express is a handbook for all stakeholders of the internet that provides information security professionals with the latest news, updates and knowledge they need to combat cyber threats.

 

Contact

For editorial queries: [email protected]

For marketing and Sales: [email protected]

 

Quick Links

  • About Us
  • Contact Us
  • Editorial Calendar
  • Careers
  • The Cyber Express by Cyble Vulnerability Disclosure Policy
  • Cyble Trust Portal

Our Address

We’re remote friendly, with office locations around the world:

San Francisco, Atlanta, Rome,
Dubai, Mumbai, Bangalore, Hyderabad,  Singapore, Jakarta, Sydney, and Melbourne

 

Headquarters:

The Cyber Express LLC
10080 North Wolfe Road, Suite SW3-200, Cupertino, CA, US 95014

 

India Office:

Cyber Express Media Network
HD-021, 4th Floor, C Wing, Building No.4. Nesco IT Park, WE Highway, Goregaon East, Mumbai, Maharashtra, India – 4000063

  • Privacy Statement
  • Terms of Use
  • Write For Us

© 2026 The Cyber Express - Cybersecurity News and Magazine.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In

Add New Playlist

No Result
View All Result
  • Magazine
  • Firewall Daily
  • Essentials
    • Regulations
    • Compliance
    • Governance
    • Policy Updates
  • Knowledge Hub
  • Features
    • Cyber Warfare
    • Espionage
    • Workforce
      • Learning & Development
  • Business
    • Startups
    • Mergers & Aquisitions
    • Partnerships
    • Appointments
    • Budgets
    • Research
      • Whitepapers
      • Sponsored Content
      • Market Reports
    • Interviews
      • Podcast
  • Events
    • Conference
    • Webinar
    • Endorsed Events
  • Advisory Board

© 2026 The Cyber Express - Cybersecurity News and Magazine.

Are you sure want to unlock this post?
Unlock left : 0
Are you sure want to cancel subscription?
-
00:00
00:00

Queue

Update Required Flash plugin
-
00:00
00:00
Do This on Telegram, Your Bank Account Will Become Zero If You Install the iOS 18 Beta, Your iPhone Could Be Hacked Cricket World Cup Ticketing Systems Under Cybersecurity Cyber Threats and Online Ticket Scams During the NBA Finals Biometric Data Security: Protecting Sensitive Information