Safety
131 posts
Responding to the next frontier of critical cyber capabilities
OpenAI is sharing preliminary cybersecurity evaluations for Astra and the steps we’re taking to strengthen safeguards and security controls.
Third-party cyber evaluations involving OpenAI models
OpenAI explains recent third-party cybersecurity evaluation incidents and outlines new safeguards to strengthen AI model testing and evaluation.
OpenAI and Hugging Face partner to address security incident during model evaluation
OpenAI and Hugging Face share early findings from a security incident during AI model evaluation, highlighting advanced cyber capabilities and lessons for de…
Safety and alignment in an era of long-horizon models
OpenAI shares lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterative depl…
Why teens deserve access to safe AI
Learn how OpenAI is making ChatGPT safer for teens with age-appropriate protections, learning tools, parental controls, and expert partnerships.
GPT-Red: Unlocking Self-Improvement for Robustness
Explore GPT-Red, OpenAI’s automated red teaming system that uses self-play to improve AI safety, alignment, and prompt injection robustness.
GPT-5.5 Bio Bug Bounty
Details about the OpenAI Bio Bounty program
Daybreak: Tools for securing every organization in the world
OpenAI introduces new Daybreak tools, including Codex Security and GPT-5.5-Cyber, to help organizations find, validate, and patch vulnerabilities at scale.
Patch the Planet: a Daybreak initiative to support open source maintainers
OpenAI introduces Patch the Planet, a Daybreak initiative helping open-source maintainers find, validate, and fix vulnerabilities with AI and expert review.
A shared playbook for trustworthy third party evaluations
OpenAI shares guidance on third-party AI evaluations, covering how to assess model capabilities, safeguards, and validity for frontier systems.
OpenAI’s Frontier Governance Framework
Explore OpenAI’s Frontier Governance Framework and how our AI safety, security, and risk practices align with emerging EU and California regulations.
Advancing content provenance for a safer, more transparent AI ecosystem
OpenAI advances AI content provenance with Content Credentials, SynthID, and a verification tool to help people identify and trust AI-generated media.
Helping ChatGPT better recognize context in sensitive conversations
Learn how new ChatGPT safety updates improve context awareness in sensitive conversations, helping detect risk over time and respond more safely.
Our response to the TanStack npm supply chain attack
OpenAI details its response to the TanStack “Mini Shai-Hulud” supply chain attack, outlines protections taken to secure systems and signing certificates, and…
Running Codex safely at OpenAI
How OpenAI runs Codex securely with sandboxing, approvals, network policies, and agent-native telemetry to support safe and compliant coding agent adoption.
Scaling Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber
OpenAI expands Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber, helping verified defenders accelerate vulnerability research and protect critical inf…
Introducing Trusted Contact in ChatGPT
Introducing Trusted Contact in ChatGPT, an optional safety feature that notifies someone you trust if serious self-harm concerns are detected.
GPT-5.5 Instant System Card
[Read on openai.com](https://openai.com/index/gpt-5-5-instant-system-card)
Advancing youth safety and wellbeing in EMEA
Explore OpenAI’s European Youth Safety Blueprint and EMEA Youth & Wellbeing Grants, advancing safe, responsible AI for teens, families, and educators.
Our commitment to community safety
Learn how OpenAI protects community safety in ChatGPT through model safeguards, misuse detection, policy enforcement, and collaboration with safety experts.
GPT-5.5 System Card
[Read on openai.com](https://openai.com/index/gpt-5-5-system-card)
Accelerating the cyber defense ecosystem that protects us all
Leading security firms and enterprises join OpenAI’s Trusted Access for Cyber, using GPT-5.4-Cyber and $10M in API grants to strengthen global cyber defense.
Trusted access for the next era of cyber defense
OpenAI expands its Trusted Access for Cyber program, introducing GPT-5.4-Cyber to vetted defenders and strengthening safeguards as AI cybersecurity capabilit…
Our response to the Axios developer tool compromise
OpenAI responds to the Axios supply chain attack by rotating macOS code signing certificates, updating apps, and confirming no user data was compromised.
Introducing the Child Safety Blueprint
Discover OpenAI’s Child Safety Blueprint—a roadmap for building AI responsibly with safeguards, age-appropriate design, and collaboration to protect and empo…
Announcing the OpenAI Safety Fellowship
A pilot program to support independent safety and alignment research and develop the next generation of talent
Introducing the OpenAI Safety Bug Bounty program
OpenAI launches a Safety Bug Bounty program to identify AI abuse and safety risks, including agentic vulnerabilities, prompt injection, and data exfiltration.
Helping developers build safer AI experiences for teens
OpenAI releases prompt-based teen safety policies for developers using gpt-oss-safeguard, helping moderate age-specific risks in AI systems.
Creating with Sora Safely
To address the novel safety challenges posed by a state-of-the-art video model as well as a new social creation platform, we’ve built Sora 2 and the Sora app…
How we monitor internal coding agents for misalignment
How OpenAI uses chain-of-thought monitoring to study misalignment in internal coding agents—analyzing real-world deployments to detect risks and strengthen A…
OpenAI Japan announces Japan Teen Safety Blueprint to put teen safety first
OpenAI Japan announces the Japan Teen Safety Blueprint, introducing stronger age protections, parental controls, and well-being safeguards for teens using ge…
Designing AI agents to resist prompt injection
How ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.
An update on our mental health-related work
OpenAI shares updates on its mental health safety work, including parental controls, trusted contacts, improved distress detection, and recent litigation dev…
Disrupting malicious uses of AI | February 2026
Our latest threat report examines how malicious actors combine AI models with websites and social platforms—and what it means for detection and defense.
Introducing Lockdown Mode and Elevated Risk labels in ChatGPT
Introducing Lockdown Mode and Elevated Risk labels in ChatGPT to help organizations defend against prompt injection and AI-driven data exfiltration.
Introducing Trusted Access for Cyber
OpenAI introduces Trusted Access for Cyber, a trust-based framework that expands access to frontier cyber capabilities while strengthening safeguards against…
Keeping your data safe when an AI agent clicks a link
Learn how OpenAI protects user data when AI agents open links, preventing URL-based data exfiltration and prompt injection with built-in safeguards.
Our approach to age prediction
ChatGPT is rolling out age prediction to estimate if accounts are under or over 18, applying safeguards for teens and refining accuracy over time.
Continuously hardening ChatGPT Atlas against prompt injection
OpenAI is strengthening ChatGPT Atlas against prompt injection attacks using automated red teaming trained with reinforcement learning. This proactive discov…
AI literacy resources for teens and parents
OpenAI shares new AI literacy resources to help teens and parents use ChatGPT thoughtfully, safely, and with confidence. The guides include expert-vetted tip…
Updating our Model Spec with teen protections
OpenAI is updating its Model Spec with new Under-18 Principles that define how ChatGPT should support teens with safe, age-appropriate guidance grounded in d…
Strengthening cyber resilience as AI capabilities advance
OpenAI is investing in stronger safeguards and defensive capabilities as AI models become more powerful in cybersecurity. We explain how we assess risk, limi…
Strengthening our safety ecosystem with external testing
OpenAI works with independent experts to evaluate frontier AI systems. Third-party testing strengthens safety, validates safeguards, and increases transparen…
Fighting the New York Times’ invasion of user privacy
OpenAI is fighting the New York Times’ demand for 20 million private ChatGPT conversations and accelerating new security and privacy protections to protect y…
Understanding prompt injections: a frontier security challenge
Prompt injections are a frontier security challenge for AI systems. Learn how these attacks work and how OpenAI is advancing research, training models, and b…
Introducing Aardvark: OpenAI’s agentic security researcher
OpenAI introduces Aardvark, an AI-powered security researcher that autonomously finds, validates, and helps fix software vulnerabilities at scale. The system…
gpt-oss-safeguard technical report
gpt-oss-safeguard-120b and gpt-oss-safeguard-20b are two open-weight reasoning models post-trained from the gpt-oss models and trained to reason from a provi…
Addendum to GPT-5 System Card: Sensitive conversations
This system card details GPT-5’s improvements in handling sensitive conversations, including new benchmarks for emotional reliance, mental health, and jailbr…
Strengthening ChatGPT’s responses in sensitive conversations
OpenAI collaborated with 170+ mental health experts to improve ChatGPT’s ability to recognize distress, respond empathetically, and guide users toward real-w…
Building towards age prediction
Learn how OpenAI is building age prediction and parental controls in ChatGPT to create safer, age-appropriate experiences for teens while supporting families…
Teen safety, freedom, and privacy
Explore OpenAI’s approach to balancing teen safety, freedom, and privacy in AI use.
Addendum to GPT-5 system card: GPT-5-Codex
This addendum to the GPT-5 system card shares a new model: GPT-5-Codex, a version of GPT-5 further optimized for agentic coding in Codex. GPT-5-Codex adjusts…
Building more helpful ChatGPT experiences for everyone
We’re partnering with experts, strengthening protections for teens with parental controls, and routing sensitive conversations to reasoning models in ChatGPT.
OpenAI and Anthropic share findings from a joint safety evaluation
OpenAI and Anthropic share findings from a first-of-its-kind joint safety evaluation, testing each other’s models for misalignment, instruction following, ha…
Helping people when they need it most
How we think about safety for users experiencing mental or emotional distress, the limits of today’s systems, and the work underway to refine them.
From hard refusals to safe-completions: toward output-centric safety training
Discover how OpenAI's new safe-completions approach in GPT-5 improves both safety and helpfulness in AI responses—moving beyond hard refusals to nuanced, out…
Estimating worst case frontier risks of open weight LLMs
In this paper, we study the worst-case frontier risks of releasing gpt-oss. We introduce malicious fine-tuning (MFT), where we attempt to elicit maximum capa…
What we’re optimizing ChatGPT for
We build ChatGPT to help you thrive in all the ways you want. Learn how we’re improving support for tough moments, have rolled out reminders to take breaks, …
Preparing for future AI risks in biology
Advanced AI can transform biology and medicine—but also raises biosecurity risks. We’re proactively assessing capabilities and implementing safeguards to pre…
Scaling security with responsible disclosure
OpenAI introduces its Outbound Coordinated Disclosure Policy to guide how it responsibly reports vulnerabilities in third-party software—emphasizing integrit…
How we’re responding to The New York Times’ data demands in order to protect user privacy
OpenAI is fighting a court order at the demands of The New York Times and plaintiffs, which involves retention of consumer ChatGPT and API user data indefini…
Addendum to OpenAI o3 and o4-mini system card: OpenAI o3 Operator
We are replacing the existing GPT-4o-based model for Operator with a version based on OpenAI o3. The API version will remain based on 4o.
Addendum to o3 and o4-mini system card: Codex
Codex is a cloud-based coding agent. Codex is powered by codex-1, a version of OpenAI o3 optimized for software engineering. codex-1 was trained using reinfo…
Security on the path to AGI
At OpenAI, we proactively adapt, including by building comprehensive security measures directly into our infrastructure and models.
Deep research System Card
This report outlines the safety work carried out prior to releasing deep research including external red teaming, frontier risk evaluations according to our …
Sharing the latest Model Spec
[Read on openai.com](https://openai.com/index/sharing-the-latest-model-spec)
Operator System Card
Drawing from OpenAI’s established safety frameworks, this document highlights our multi-layered approach, including model and product mitigations we’ve imple…
Deliberative alignment: reasoning enables safer language models
Deliberative alignment: reasoning enables safer language models Introducing our new alignment strategy for o1 models, which are directly taught safety specif…
Sora System Card
Sora is OpenAI’s video generation model, designed to take text, image, and video inputs and generate a new video as an output. Sora builds on learnings from …
An update on our safety & security practices
An update on our safety & security practices
Disrupting a covert Iranian influence operation
[Read on openai.com](https://openai.com/index/disrupting-a-covert-iranian-influence-operation)
GPT-4o System Card
[Read on openai.com](https://openai.com/index/gpt-4o-system-card)
Finding GPT-4’s mistakes with GPT-4
CriticGPT, a model based on GPT-4, writes critiques of ChatGPT responses to help human trainers spot mistakes during RLHF
Expanding on how Voice Engine works and our safety research
Exploring the technology behind our text-to-speech model.
Disrupting deceptive uses of AI by covert influence operations
We’ve terminated accounts linked to covert influence operations; no significant audience increase due to our services.
OpenAI safety practices
Artificial general intelligence has the potential to benefit nearly every aspect of our lives—so it must be developed and deployed responsibly.
Our approach to data and AI
Just over a year after launching ChatGPT, AI is changing how we live, work and learn. It’s also raised important conversations about data in the age of AI. M…
OpenAI’s commitment to child safety: adopting safety by design principles
[Read on openai.com](https://openai.com/index/child-safety-adopting-sbd-principles)
Disrupting malicious uses of AI by state-affiliated threat actors
[Read on openai.com](https://openai.com/index/disrupting-malicious-uses-of-ai-by-state-affiliated-threat-actors)
Democratic inputs to AI grant program: lessons learned and implementation plans
We funded 10 teams from around the world to design ideas and tools to collectively govern AI. We summarize the innovations, outline our learnings, and call f…
How OpenAI is approaching 2024 worldwide elections
We’re working to prevent abuse, provide transparency on AI-generated content, and improve access to accurate voting information.
Practices for Governing Agentic AI Systems
[Read on openai.com](https://openai.com/index/practices-for-governing-agentic-ai-systems)
Superalignment Fast Grants
We’re launching $10M in grants to support technical research towards the alignment and safety of superhuman AI systems, including weak-to-strong generalizati…
Weak-to-strong generalization
We present a new research direction for superalignment, together with promising initial results: can we leverage the generalization properties of deep learni…
Frontier risk and preparedness
To support the safety of highly-capable AI systems, we are developing our approach to catastrophic risk preparedness, including building a Preparedness team …
DALL·E 3 system card
[Read on openai.com](https://openai.com/index/dall-e-3-system-card)
GPT-4V(ision) system card
[Read on openai.com](https://openai.com/index/gpt-4v-system-card)
OpenAI Red Teaming Network
We’re announcing an open call for the OpenAI Red Teaming Network and invite domain experts interested in improving the safety of OpenAI’s models to join our …
Using GPT-4 for content moderation
We use GPT-4 for content policy development and content moderation decisions, enabling more consistent labeling, a faster feedback loop for policy refinement…
Confidence-Building Measures for Artificial Intelligence: Workshop proceedings
[Read on openai.com](https://openai.com/index/confidence-building-measures-for-artificial-intelligence)
Frontier Model Forum
We’re forming a new industry body to promote the safe and responsible development of frontier AI systems: advancing AI safety research, identifying best prac…
Moving AI governance forward
OpenAI and other leading labs reinforce AI safety, security and trustworthiness through voluntary commitments.
Frontier AI regulation: Managing emerging risks to public safety
[Read on openai.com](https://openai.com/index/frontier-ai-regulation)
Insights from global conversations
We are sharing what we learned from our conversations across 22 countries, and how we will be incorporating those insights moving forward.
Governance of superintelligence
Now is a good time to start thinking about the governance of superintelligence—future AI systems dramatically more capable than even AGI.
Language models can explain neurons in language models
We use GPT-4 to automatically write explanations for the behavior of neurons in large language models and to score those explanations. We release a dataset o…
Announcing OpenAI’s Bug Bounty Program
This initiative is essential to our commitment to develop safe and advanced AI. As we create technology and services that are secure, reliable, and trustwort…
Our approach to AI safety
Ensuring that AI systems are built, deployed, and used safely is critical to our mission.
Planning for AGI and beyond
Our mission is to ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.
How should AI systems behave, and who should decide?
We’re clarifying how ChatGPT’s behavior is shaped and our plans for improving that behavior, allowing more user customization, and getting more public input …
Forecasting potential misuses of language models for disinformation campaigns and how to reduce risk
OpenAI researchers collaborated with Georgetown University’s Center for Security and Emerging Technology and the Stanford Internet Observatory to investigate…
Our approach to alignment research
We are improving our AI systems’ ability to learn from human feedback and to assist humans at evaluating AI. Our goal is to build a sufficiently aligned AI s…
A hazard analysis framework for code synthesis large language models
[Read on openai.com](https://openai.com/index/a-hazard-analysis-framework-for-code-synthesis-large-language-models)
AI-written critiques help humans notice flaws
We trained “critique-writing” models to describe flaws in summaries. Human evaluators find flaws in summaries much more often when shown our model’s critique…
Best practices for deploying language models
Cohere, OpenAI, and AI21 Labs have developed a preliminary set of best practices applicable to any organization developing or deploying large language models.
Measuring Goodhart’s law
Goodhart’s law famously says: “When a measure becomes a target, it ceases to be a good measure.” Although originally from economics, it’s something we have t…
Economic impacts research at OpenAI
Call for expressions of interest to study the economic impacts of large language models.
Lessons learned on language model safety and misuse
We describe our latest thinking in the hope of helping other AI developers address safety and misuse of deployed models.
Aligning language models to follow instructions
We’ve trained language models that are much better at following user intentions than GPT-3 while also making them more truthful and less toxic, using techniq…
Summarizing books with human feedback
Scaling human oversight of AI systems for tasks that are difficult to evaluate.
Improving language model behavior by training on a curated dataset
Our latest research finds we can improve language model behavior with respect to specific behavioral values by fine-tuning on a small, curated dataset.
Learning to summarize with human feedback
We’ve applied reinforcement learning from human feedback to train language models that are better at summarization.
Benchmarking safe exploration in deep reinforcement learning
[Read on openai.com](https://openai.com/index/benchmarking-safe-exploration-in-deep-reinforcement-learning)
Safety Gym
We’re releasing Safety Gym, a suite of environments and tools for measuring progress towards reinforcement learning agents that respect safety constraints wh…
Fine-tuning GPT-2 from human preferences
We’ve fine-tuned the 774M parameter GPT-2 language model using human feedback for various tasks, successfully matching the preferences of the external human …
Testing robustness against unforeseen adversaries
We’ve developed a method to assess whether a neural network classifier can reliably defend against adversarial attacks not seen during training. Our method y…
Why responsible AI development needs cooperation on safety
We’ve written a policy research paper identifying four strategies that can be used today to improve the likelihood of long-term industry cooperation on safet…
Transfer of adversarial robustness between perturbation types
[Read on openai.com](https://openai.com/index/transfer-of-adversarial-robustness-between-perturbation-types)
Introducing Activation Atlases
We’ve created activation atlases (in collaboration with Google researchers), a new technique for visualizing what interactions between neurons can represent.…
AI safety needs social scientists
We’ve written a paper arguing that long-term AI safety research needs social scientists to ensure AI alignment algorithms succeed when actual humans are invo…
Learning complex goals with iterated amplification
We’re proposing an AI safety technique called iterated amplification that lets us specify complicated behaviors and goals that are beyond human scale, by dem…
Improving language understanding with unsupervised learning
We’ve obtained state-of-the-art results on a suite of diverse language tasks with a scalable, task-agnostic system, which we’re also releasing. Our approach …
AI safety via debate
We’re proposing an AI safety technique which trains agents to debate topics with one another, using a human to judge who wins.
Preparing for malicious uses of AI
We’ve co-authored a paper that forecasts how malicious actors could misuse AI technology, and potential ways we can prevent and mitigate these threats. This …
Learning from human preferences
One step towards building safe AI systems is to remove the need for humans to write goal functions, since using a simple proxy for a complex goal, or getting…
Attacking machine learning with adversarial examples
Adversarial examples are inputs to machine learning models that an attacker has intentionally designed to cause the model to make a mistake; they’re like opt…
Adversarial attacks on neural network policies
[Read on openai.com](https://openai.com/index/adversarial-attacks-on-neural-network-policies)
Faulty reward functions in the wild
Reinforcement learning algorithms can break in surprising, counterintuitive ways. In this post we’ll explore one failure mode, which is where you misspecify …
Semi-supervised knowledge transfer for deep learning from private training data
[Read on openai.com](https://openai.com/index/semi-supervised-knowledge-transfer-for-deep-learning-from-private-training-data)
Concrete AI safety problems
We (along with researchers from Berkeley and Stanford) are co-authors on today’s paper led by Google Brain researchers, Concrete Problems in AI Safety. The p…
Adversarial training methods for semi-supervised text classification
[Read on openai.com](https://openai.com/index/adversarial-training-methods-for-semi-supervised-text-classification)
