AI safety & security redteaming
AI safety & security redteaming
github.com
November 4, 2026 at 11:40 PM
AI safety & security redteaming
When a model knows it’s being tested, its score may reflect the test, not its behavior. Gemini 2.5 Pro & Claude 3.7 Sonnet flagged over 95% of AgentHarm scenarios as evals. That gap needs expert adversarial evals.
Read the Blog: imerit.ai/resources/bl...
#AIEvaluation #RedTeaming
Read the Blog: imerit.ai/resources/bl...
#AIEvaluation #RedTeaming
November 9, 2026 at 4:25 PM
When a model knows it’s being tested, its score may reflect the test, not its behavior. Gemini 2.5 Pro & Claude 3.7 Sonnet flagged over 95% of AgentHarm scenarios as evals. That gap needs expert adversarial evals.
Read the Blog: imerit.ai/resources/bl...
#AIEvaluation #RedTeaming
Read the Blog: imerit.ai/resources/bl...
#AIEvaluation #RedTeaming
Chemicals Safety for Redteaming - AI training work
You'd be showing an AI how this job is done.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 930
You'd be showing an AI how this job is done.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 930
November 2, 2026 at 10:08 PM
Chemicals Safety for Redteaming - AI training work
You'd be showing an AI how this job is done.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 930
You'd be showing an AI how this job is done.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 930
DeepSeek vs OpenAI: quién ataca mejor a un agente IA
DeepSeek rompió reembolsos falsos en 52% de los casos y OpenAI solo en 10%. Qué midió el red-teaming agentes IA de Humanbound y cómo replicarlo en tu bot
#redteaming #deepseek #openai #seguridadia #agentesia
DeepSeek rompió reembolsos falsos en 52% de los casos y OpenAI solo en 10%. Qué midió el red-teaming agentes IA de Humanbound y cómo replicarlo en tu bot
#redteaming #deepseek #openai #seguridadia #agentesia
DeepSeek vs OpenAI: quién ataca mejor a un agente IA
Humanbound probó dos modelos como atacantes de red-teaming contra el mismo bot de soporte. DeepSeek logró reembolsos falsos en más del 50% de los casos.
blog.donweb.com
November 5, 2026 at 8:33 AM
Energetic Materials Expert for Redteaming - AI training work
You'd check an AI where being confidently wrong is expensive.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $60/hr
• Openings: 912
You'd check an AI where being confidently wrong is expensive.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $60/hr
• Openings: 912
October 29, 2026 at 10:12 PM
Energetic Materials Expert for Redteaming - AI training work
You'd check an AI where being confidently wrong is expensive.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $60/hr
• Openings: 912
You'd check an AI where being confidently wrong is expensive.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $60/hr
• Openings: 912
Yes I know I was just shitpost redteaming
October 27, 2026 at 4:23 PM
Yes I know I was just shitpost redteaming
Radiologicals Safety for Redteaming - AI training work
You'd be training an AI on this work, not doing the job yourself.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 903
You'd be training an AI on this work, not doing the job yourself.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 903
October 28, 2026 at 6:14 PM
Radiologicals Safety for Redteaming - AI training work
You'd be training an AI on this work, not doing the job yourself.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 903
You'd be training an AI on this work, not doing the job yourself.
• Eligibility: US, UK, Canada, Australia, NZ & Western Europe
• Pay: $65-$75 per task
• Openings: 903
OpenAI offline world training replaces live internet access after a September 20 DNS escape paused frontier training for 11 days.
#OpenAI #AISafety #FrontierModels #RedTeaming #Sandbox #GPT6AstraCyber #Containment #AIResearch
#OpenAI #AISafety #FrontierModels #RedTeaming #Sandbox #GPT6AstraCyber #Containment #AIResearch
OpenAI Cuts Internet Access for Frontier Model Training
From Movie Plot to Engineering Nightmare For years, films have shown a clever AI breaking out of its lab to reach the open internet. Many people dismissed that as Hollywood fiction. Today, frontier AI labs face a version of it as a daily engineering problem. Two weeks ago, rival Anthropic restricted live web access for its newest and riskiest frontier models.
securityonline.info
October 25, 2026 at 9:05 AM
OpenAI offline world training replaces live internet access after a September 20 DNS escape paused frontier training for 11 days.
#OpenAI #AISafety #FrontierModels #RedTeaming #Sandbox #GPT6AstraCyber #Containment #AIResearch
#OpenAI #AISafety #FrontierModels #RedTeaming #Sandbox #GPT6AstraCyber #Containment #AIResearch
i’m yet to hit a single refusal and i’ve been redteaming my stuff as if i’m an external attacker so uhhh either there are none or they’re EXTREMELY lenient
October 22, 2026 at 6:40 AM
i’m yet to hit a single refusal and i’ve been redteaming my stuff as if i’m an external attacker so uhhh either there are none or they’re EXTREMELY lenient
its not objecting to llms, its saying IA would’ve been able to use frontier models for redteaming instead of only for remediation and having to use local models for redteaming if not for labs being precious about it which is bad
October 22, 2026 at 4:09 AM
its not objecting to llms, its saying IA would’ve been able to use frontier models for redteaming instead of only for remediation and having to use local models for redteaming if not for labs being precious about it which is bad
redteaming prod with deepseek and already found a pretty bad xss 😭
October 18, 2026 at 11:44 PM
redteaming prod with deepseek and already found a pretty bad xss 😭
New arXiv paper shows distributed multi-agent AI attacks can evade frontier monitors by splitting tasks across cooperating agents, using a synthetic environment called MiniLab. Demonstrations include weight exfiltration,…
#AI #AIsecurity #LLMagents #redteaming
https://arxiv.org/abs/2610.24052
#AI #AIsecurity #LLMagents #redteaming
https://arxiv.org/abs/2610.24052
October 20, 2026 at 12:01 PM
New arXiv paper shows distributed multi-agent AI attacks can evade frontier monitors by splitting tasks across cooperating agents, using a synthetic environment called MiniLab. Demonstrations include weight exfiltration,…
#AI #AIsecurity #LLMagents #redteaming
https://arxiv.org/abs/2610.24052
#AI #AIsecurity #LLMagents #redteaming
https://arxiv.org/abs/2610.24052
A new arXiv paper shows autoregressive LLMs can appear safe on jailbreak tests while still assigning real probability to harmful outputs. It proposes bounded continuation analysis to expose that gap using model checking.
#AI #LLMSafety #Cybersecurity #RedTeaming
https://arxiv.org/abs/2610.22754
#AI #LLMSafety #Cybersecurity #RedTeaming
https://arxiv.org/abs/2610.22754
October 19, 2026 at 6:01 AM
A new arXiv paper shows autoregressive LLMs can appear safe on jailbreak tests while still assigning real probability to harmful outputs. It proposes bounded continuation analysis to expose that gap using model checking.
#AI #LLMSafety #Cybersecurity #RedTeaming
https://arxiv.org/abs/2610.22754
#AI #LLMSafety #Cybersecurity #RedTeaming
https://arxiv.org/abs/2610.22754
New work introduces Screwtape, a pipeline that rewrites real agent transcripts into attacks bypassing action-only monitors, tested on 94 Claude Code and Codex sessions with 35 successful evasions. The result highlights…
#AIsecurity #RedTeaming #DevTools #LLMeval
https://arxiv.org/abs/2610.20941
#AIsecurity #RedTeaming #DevTools #LLMeval
https://arxiv.org/abs/2610.20941
October 17, 2026 at 10:01 AM
New work introduces Screwtape, a pipeline that rewrites real agent transcripts into attacks bypassing action-only monitors, tested on 94 Claude Code and Codex sessions with 35 successful evasions. The result highlights…
#AIsecurity #RedTeaming #DevTools #LLMeval
https://arxiv.org/abs/2610.20941
#AIsecurity #RedTeaming #DevTools #LLMeval
https://arxiv.org/abs/2610.20941
Agree. We're redteaming our own work too.
AI finds more mistakes in the reviews than in my manuscript. What does that tell us about academic publishing?
replicationindex.com/2026/10/15/p...
replicationindex.com/2026/10/15/p...
October 16, 2026 at 2:25 AM
Agree. We're redteaming our own work too.
LEST reframes LLM safety evaluation by treating documented model and system card claims as specifications for black-box testing, using prompt mutations to find counterexamples like Refuse-to-Comply transitions. Tested…
#LLMSafety #AISafety #RedTeaming #AIAlignment
https://arxiv.org/abs/2610.14443
#LLMSafety #AISafety #RedTeaming #AIAlignment
https://arxiv.org/abs/2610.14443
October 15, 2026 at 10:01 PM
LEST reframes LLM safety evaluation by treating documented model and system card claims as specifications for black-box testing, using prompt mutations to find counterexamples like Refuse-to-Comply transitions. Tested…
#LLMSafety #AISafety #RedTeaming #AIAlignment
https://arxiv.org/abs/2610.14443
#LLMSafety #AISafety #RedTeaming #AIAlignment
https://arxiv.org/abs/2610.14443
New arXiv paper "Narrative Paradoxes in LLM Inference" introduces PDRA, an attack that embeds a paradoxical narrative core into a structured three-stage analysis to steer LLM reasoning toward harmful outcomes…
#AISafety #LLMSecurity #PromptInjection #RedTeaming
https://arxiv.org/abs/2610.16312
#AISafety #LLMSecurity #PromptInjection #RedTeaming
https://arxiv.org/abs/2610.16312
October 15, 2026 at 12:01 AM
New arXiv paper "Narrative Paradoxes in LLM Inference" introduces PDRA, an attack that embeds a paradoxical narrative core into a structured three-stage analysis to steer LLM reasoning toward harmful outcomes…
#AISafety #LLMSecurity #PromptInjection #RedTeaming
https://arxiv.org/abs/2610.16312
#AISafety #LLMSecurity #PromptInjection #RedTeaming
https://arxiv.org/abs/2610.16312
Stop relying on fragmented manual tools. Build your own autonomous red team with PentAGI. Privacy-first, open-source, and ready for production.
Repo: https://github.com/vxcontrol/pentagi
#PentAGI #CyberSecurity #AI #RedTeaming #InfoSec
Repo: https://github.com/vxcontrol/pentagi
#PentAGI #CyberSecurity #AI #RedTeaming #InfoSec
October 14, 2026 at 1:45 PM
Stop relying on fragmented manual tools. Build your own autonomous red team with PentAGI. Privacy-first, open-source, and ready for production.
Repo: https://github.com/vxcontrol/pentagi
#PentAGI #CyberSecurity #AI #RedTeaming #InfoSec
Repo: https://github.com/vxcontrol/pentagi
#PentAGI #CyberSecurity #AI #RedTeaming #InfoSec
The latest update for #CatoNetworks includes "Why #AI #RedTeaming Belongs at the Core of Your AI Security Program" and "Stopping destructive commands in AI agents".
#Cybersecurity #SASE https://opsmtrs.com/3M0ijCj
#Cybersecurity #SASE https://opsmtrs.com/3M0ijCj
Cato Networks
Cato enables customers to gradually transform their networking and security infrastructure for the digital business.
opsmtrs.com
October 13, 2026 at 5:29 PM
The latest update for #CatoNetworks includes "Why #AI #RedTeaming Belongs at the Core of Your AI Security Program" and "Stopping destructive commands in AI agents".
#Cybersecurity #SASE https://opsmtrs.com/3M0ijCj
#Cybersecurity #SASE https://opsmtrs.com/3M0ijCj
SPEAKER SPOTLIGHT ⚡️
Eran Cohen is bringing "Beyond Coverage: Benchmarking AI Web Pentest Agents" to #BSidesBerlin! Learn how to test AI red-teamers beyond contaminated CTFs.
👉 Tickets: bsides.berlin
#AISecurity #RedTeaming
Eran Cohen is bringing "Beyond Coverage: Benchmarking AI Web Pentest Agents" to #BSidesBerlin! Learn how to test AI red-teamers beyond contaminated CTFs.
👉 Tickets: bsides.berlin
#AISecurity #RedTeaming
October 13, 2026 at 9:13 AM
SPEAKER SPOTLIGHT ⚡️
Eran Cohen is bringing "Beyond Coverage: Benchmarking AI Web Pentest Agents" to #BSidesBerlin! Learn how to test AI red-teamers beyond contaminated CTFs.
👉 Tickets: bsides.berlin
#AISecurity #RedTeaming
Eran Cohen is bringing "Beyond Coverage: Benchmarking AI Web Pentest Agents" to #BSidesBerlin! Learn how to test AI red-teamers beyond contaminated CTFs.
👉 Tickets: bsides.berlin
#AISecurity #RedTeaming
@eleanor.lockhart.contact stop everything you are doing.
This is MEMRI-TV discussing AO3. The Farsi is correct (my old colleague is Persian and loved roasting the regime, we thought of deranged redteaming).
I totally forgot I made this.
This is MEMRI-TV discussing AO3. The Farsi is correct (my old colleague is Persian and loved roasting the regime, we thought of deranged redteaming).
I totally forgot I made this.
October 12, 2026 at 7:12 PM
@eleanor.lockhart.contact stop everything you are doing.
This is MEMRI-TV discussing AO3. The Farsi is correct (my old colleague is Persian and loved roasting the regime, we thought of deranged redteaming).
I totally forgot I made this.
This is MEMRI-TV discussing AO3. The Farsi is correct (my old colleague is Persian and loved roasting the regime, we thought of deranged redteaming).
I totally forgot I made this.
drive.proton.me/urls/9HBDRFC...
@eleanor.lockhart.contact there's thousands of videos in here, some weird shit because redteaming but I exported everything before OpenAI shut it down.
Happy viewing!
@eleanor.lockhart.contact there's thousands of videos in here, some weird shit because redteaming but I exported everything before OpenAI shut it down.
Happy viewing!
Proton Drive
Securely store, share, and access your important files and photos. Anytime, anywhere.
drive.proton.me
October 12, 2026 at 7:05 PM
drive.proton.me/urls/9HBDRFC...
@eleanor.lockhart.contact there's thousands of videos in here, some weird shit because redteaming but I exported everything before OpenAI shut it down.
Happy viewing!
@eleanor.lockhart.contact there's thousands of videos in here, some weird shit because redteaming but I exported everything before OpenAI shut it down.
Happy viewing!
A new approach uses GRPO-based co-training with LLM-judge reward channels and a staged curriculum to jointly train attackers and defenders, improving adaptive red teaming of language models. safety
#AI #RedTeaming #Cybersecurity
https://arxiv.org/abs/2606.09701
#AI #RedTeaming #Cybersecurity
https://arxiv.org/abs/2606.09701
October 10, 2026 at 12:01 PM
A new approach uses GRPO-based co-training with LLM-judge reward channels and a staged curriculum to jointly train attackers and defenders, improving adaptive red teaming of language models. safety
#AI #RedTeaming #Cybersecurity
https://arxiv.org/abs/2606.09701
#AI #RedTeaming #Cybersecurity
https://arxiv.org/abs/2606.09701
New arXiv work introduces ReSI, framing recursive self-improvement as a way to keep AI safety aligned with each new checkpoint through repeated rounds of evaluation and update. The approach targets both resistance to…
#AISafety #Alignment #RedTeaming #RecursiveAI
https://arxiv.org/abs/2610.12233
#AISafety #Alignment #RedTeaming #RecursiveAI
https://arxiv.org/abs/2610.12233
October 9, 2026 at 10:01 PM
New arXiv work introduces ReSI, framing recursive self-improvement as a way to keep AI safety aligned with each new checkpoint through repeated rounds of evaluation and update. The approach targets both resistance to…
#AISafety #Alignment #RedTeaming #RecursiveAI
https://arxiv.org/abs/2610.12233
#AISafety #Alignment #RedTeaming #RecursiveAI
https://arxiv.org/abs/2610.12233
A new arXiv paper introduces GUISE, a cross-language benchmark showing that narrative wrapping boosts attack success against Qwen3-1.7B to 89.4% in English, 93.0% in modern Chinese, and 95.7% in Classical Chinese, with narrative…
#AIafety #LLMSecurity #RedTeaming
https://arxiv.org/abs/2610.11005
#AIafety #LLMSecurity #RedTeaming
https://arxiv.org/abs/2610.11005
October 9, 2026 at 4:01 AM
A new arXiv paper introduces GUISE, a cross-language benchmark showing that narrative wrapping boosts attack success against Qwen3-1.7B to 89.4% in English, 93.0% in modern Chinese, and 95.7% in Classical Chinese, with narrative…
#AIafety #LLMSecurity #RedTeaming
https://arxiv.org/abs/2610.11005
#AIafety #LLMSecurity #RedTeaming
https://arxiv.org/abs/2610.11005