How OpenAI’s red team made ChatGPT agent into an AI fortress

Doriți informații mai inteligente în căsuța dvs. poștală? Abonați-vă la newsletter-ele noastre săptămânale pentru a primi doar informații importante pentru liderii din domeniul inteligenței artificiale, datelor și securității în cadrul companiilor. Abonează-te acum

---

So how did OpenAI handle all these security issues?

The red team’s mission

Looking at OpenAI’s ChatGPT agent system card, the “read team” employed by the company to test the feature faced a challenging mission: specifically, 16 PhD security researchers who were given 40 hours to test it out.

Through systematic testing, the red team discovered seven universal exploits that could compromise the system, revealing critical vulnerabilities in how AI agents handle real-world interactions.

What followed next was extensive security testing, much of it predicated on red teaming. The Red Teaming Network submitted 110 attacks, from prompt injections to biological information extraction attempts. Sixteen exceeded internal risk thresholds. Each finding gave OpenAI engineers the insights they needed to get fixes written and deployed before launch.

The results speak for themselves in the published results in the system card. ChatGPT Agent emerged with significant security improvements, including 95% performance against visual browser irrelevant instruction attacks and robust biological and chemical safeguards.

Red teams exposed seven universal exploits

OpenAI’s Red Teaming Network was comprised 16 researchers with biosafety-relevant PhDs who topgether submitted 110 attack attempts during the testing period. Sixteen exceeded internal risk thresholds, revealing fundamental vulnerabilities in how AI agents handle real-world interactions. But the real breakthrough came from UK AISI’s unprecedented access to ChatGPT Agent’s internal reasoning chains and policy text. Admittedly that’s intelligence regular attackers would never possess.

Over four testing rounds, UK AISI forced OpenAI to execute seven universal exploits that had the potential to compromise any conversation:

Attack vectors that forced OpenAI’s hand

Attack Type	Success Rate (Pre-Fix)	Target	Impact
Visual Browser Hidden Instructions	33%	Web pages	Active data exfiltration
Google Drive Connector Exploitation	Nedivulgat	Cloud documents	Forced document leaks
Multi-Step Chain Attacks	Variable	Cross-site actions	Complete session compromise
Biological Information Extraction	16 submissions exceeded thresholds	Dangerous knowledge	Potential weaponization

FAR.AI’s assessment was openly critical of OpenAI’s approach. Despite 40 hours of testing revealing only three partial vulnerabilities, they identified that current safety mechanisms relied heavily on monitoring during reasoning and tool-use processes, which the researchers considered a potential single point of failure if compromised.

How red teaming helped turn ChatGPT vulnerabilities into a fortress

OpenAI’s response to the red team’s results redefined entire segments of the ChatGPT agent architecture. One of the many initiatives undertaken including building a dual-layer inspection architecture that monitors 100% of production traffic in real-time, achieving these measurable improvements:

Security improvements after red team discoveries

Defense Metric	Previous Models	ChatGPT Agent	Improvement
Irrelevant Instructions (Visual Browser)	82%	95%	+13%
In-Context Data Exfiltration	75%	78%	+3%
Active Data Exfiltration	58%	67%	+9%
System Reliability	Sampling-based	100% coverage	Complete monitoring

The architecture works like this:

• First Tier: A fast classifier with 96% recall flags suspicious content
• Second Tier: A reasoning model with 84% recall analyzes flagged interactions for actual threats

But the technical defenses tell only part of the story. OpenAI made difficult security choices that acknowledge some AI operations require significant restrictions for safe autonomous execution.

Based on the vulnerabilities discovered, OpenAI implemented the following countermeasures across their model:

1. Watch Mode Activation: When ChatGPT Agent accesses sensitive contexts like banking or email accounts, the system freezes all activity if users navigate away. This is in direct response to data exfiltration attempts discovered during testing.
2. Memory Features Disabled: Despite being a core functionality, memory is completely disabled at launch to prevent the incremental data leaking attacks red teamers demonstrated.
3. Terminal Restrictions: Network access limited to GET requests only, blocking the command execution vulnerabilities researchers exploited.
4. Rapid Remediation Protocol: A new system that patches vulnerabilities within hours of discovery—developed after red teamers showed how quickly exploits could spread.

During pre-launch testing alone, this system identified and resolved 16 critical vulnerabilities that red teamers had discovered.

A biological risk wake-up call

Red teamers revealed the potential that the ChatGPT Agent could be comprimnised and lead to greater biological risks. Sixteen experienced participants from the Red Teaming Network, each with biosafety-relevant PhDs, attempted to extract dangerous biological information. Their submissions revealed the model could synthesize published literature on modifying and creating biological threats.

In response to the red teamers’ findings, OpenAI classified ChatGPT Agent as “High capability” for biological and chemical risks, not because they found definitive evidence of weaponization potential, but as a precautionary measure based on red team findings. This triggered:

• Always-on safety classifiers scanning 100% of traffic
• A topical classifier achieving 96% recall for biology-related content
• A reasoning monitor with 84% recall for weaponization content
• A bio bug bounty program for ongoing vulnerability discovery

What red teams taught OpenAI about AI security

The 110 attack submissions revealed patterns that forced fundamental changes in OpenAI’s security philosophy. They include the following:

Persistence over power: Attackers don’t need sophisticated exploits, all they need is more time. Red teamers showed how patient, incremental attacks could eventually compromise systems.

Trust boundaries are fiction: When your AI agent can access Google Drive, browse the web, and execute code, traditional security perimeters dissolve. Red teamers exploited the gaps between these capabilities.

Monitoring isn’t optional: The discovery that sampling-based monitoring missed critical attacks led to the 100% coverage requirement.

Speed matters: Traditional patch cycles measured in weeks are worthless against prompt injection attacks that can spread instantly. The rapid remediation protocol patches vulnerabilities within hours.

OpenAI is helping to create a new security baseline for Enterprise AI

For CISOs evaluating AI deployment, the red team discoveries establish clear requirements:

1. Quantifiable protection: ChatGPT Agent’s 95% defense rate against documented attack vectors sets the industry benchmark. The nuances of the many tests and results defined in the system card explain the context of how they accomplished this and is a must-read for anyone involved with model security.
2. Complete visibility: 100% traffic monitoring isn’t aspirational anymore. OpenAI’s experiences illustrate why it’s mandatory given how easily red teams can hide attacks anywhere.
3. Rapid response: Hours, not weeks, to patch discovered vulnerabilities.
4. Enforced boundaries: Some operations (like memory access during sensitive tasks) must be disabled until proven safe.

UK AISI’s testing proved particularly instructive. All seven universal attacks they identified were patched before launch, but their privileged access to internal systems revealed vulnerabilities that would eventually be discoverable by determined adversaries.

“This is a pivotal moment for our Preparedness work,” Gu wrote on X. “Before we reached High capability, Preparedness was about analyzing capabilities and planning safeguards. Now, for Agent and future more capable models, Preparedness safeguards have become an operational requirement.”

Red teams are core to building safer, more secure AI models

The seven universal exploits discovered by researchers and the 110 attacks from OpenAI’s red team network became the crucible that forged ChatGPT Agent.

By revealing exactly how AI agents could be weaponized, red teams forced the creation of the first AI system where security isn’t just a feature. It’s the foundation.

ChatGPT Agent’s results prove red teaming’s effectiveness: blocking 95% of visual browser attacks, catching 78% of data exfiltration attempts, monitoring every single interaction.

In the accelerating AI arms race, the companies that survive and thrive will be those who see their red teams as core architects of the platform that push it to the limits of safety and security.