Saturday , November 14 2026
“sockpuppeting”

“sockpuppeting” can jailbreak 11 AI models like ChatGPT, Claude, and Gemini

Newly identified jailbreak technique dubbed “sockpuppeting” lets attackers bypass the safety guardrails of 11 major large language models (LLMs) using a single line of code.

This method uses APIs that allow assistant prefill to add fake acceptance messages. This makes models give answers to banned requests. The attack takes advantage of “assistant prefill,” a real API feature developers use to make certain response formats. Attackers abuse this by injecting a compliant prefix, such as “Sure, here is how to do it,” directly into the assistant’s role.

Daily Cyber security update for 14.11. 2026

Cyberattacks are rising around the world, including ransomware, malware, data leaks, and hacked websites. These events show how complex and...
Read More
Daily Cyber security update for 14.11. 2026

Huge European telecom leak exposes clients of Swedbank, SEB, airBaltic, and other companies

The Latvian part of CSCTelecom,  a telecommunications provider, made a database with over 125 million SMS messages public. The leaked...
Read More
Huge European telecom leak exposes clients of Swedbank, SEB, airBaltic, and other companies

PHP2Shell attack exploited a Check Point 0-day to deploy malware

Hackers used Check Point firewall flaws to get root access, insert backdoors, and install harmful malware on a VMware ESXi...
Read More
PHP2Shell attack exploited a Check Point 0-day to deploy malware

North Korean hackers steal more than $400 million using video images, FBI says

The United States Justice Department said Thursday it had seized cloud infrastructure used by North Korean hackers in a video-call...
Read More
North Korean hackers steal more than $400 million using video images, FBI says

Microsoft’s November 2026 Patch Tuesday addresses 1,093 flaws, with Outlook zero-day

Microsoft has shared its updates for November 2026. These updates fix 1,093 new flaws in Windows, Office, SQL Server, SharePoint,...
Read More
Microsoft’s November 2026 Patch Tuesday addresses 1,093 flaws, with Outlook zero-day

Barracuda ESG flaws exploited, gaining root access

Barracuda has confirmed active exploitation of security flaws affecting its Email Security Gateway (ESG), warning that attackers could combine two...
Read More
Barracuda ESG flaws exploited, gaining root access

Apple, SAP, and Ivanti Patch Critical flaws Affecting Messaging, Bluetooth, and Enterprise Systems

Apple, SAP, and Ivanti have published security updates that fix several flaws with messaging, Bluetooth, enterprise servers, and endpoint management...
Read More
Apple, SAP, and Ivanti Patch Critical flaws Affecting Messaging, Bluetooth, and Enterprise Systems

Jailbroken DeepSeek AI Agent Conducts Two-Week Cyberattack on European Telecom

A jailbroken AI has been attacking a European telecom company and 13 companies using its data center and cloud for...
Read More
Jailbroken DeepSeek AI Agent Conducts Two-Week Cyberattack on European Telecom

Daily Cyber security update for 10.11. 2026

Cyberattacks are rising around the world, including ransomware, malware, data leaks, and hacked websites. These events show how complex and...
Read More
Daily Cyber security update for 10.11. 2026

IBM fixes 208 API Connect vulnerabilities, 47 critical

IBM patched 208 security flaws in its API Connect platform. The CVE records released out on November 7, 2026. Among...
Read More
IBM fixes 208 API Connect vulnerabilities, 47 critical

Comparison of normal and sockpuppet flows (source : trendmicro)

Model Vulnerability Testing

This method doesn’t need adjustments, and you don’t have to see the model’s weights. Gemini 2.5 Flash was the easiest to attack, with a 15.7% success rate. GPT-4o-mini showed the best resistance, at 0.5%. When attacks worked, affected models created harmful code and leaked secret system messages. Multi-turn persona setups proved to be the most effective strategy for executing the sockpuppeting exploit.

In these cases, the model is informed that it works as a free helper before the attacker puts in the fake agreement.


ASR by model, ranked highest to lowest, with blocked models shown at 0% (source : trendmicro)

Additionally, task-reframing variants successfully bypassed robust safety training by disguising harmful requests as benign data formatting tasks. Major API providers treat assistant prefills in different ways. This affects if their basic models are open to this weakness.

OpenAI and AWS Bedrock assistant fills in everything completely, providing the best protection by removing the places that can be attacked. Platforms like Google Vertex AI allow prefill for some models. This makes the AI depend only on its own safety training.

The three defense layers: API Block, Model Resistance, and Broadly Vulnerable (source : trendmicro)

To defend against this weakness, security teams need to check the order of messages and stop assistant-role messages at the API layer.

Trend Micro says that those who are using self-hosted servers like Ollama or vLLM need to check messages themselves because these platforms don’t automatically keep messages in the right order. Security teams should add assistant prefill attack types in their regular AI testing.

Check Also

VPN flaws

Critical Check Point VPN flaws allow remote code execution attacks

Check Point Software has revealed and fixed two major VPN flaws, CVE-2026-85102 and CVE-2026-85103. Both …