OpenAI

Creator of ChatGPT and GPT-4; leading frontier AI lab focused on artificial general intelligence.

Safety Documents

✅

Responsible Scaling Policy

Preparedness Framework. Beta v1 published November 2023; v2 updated April 15, 2025. Covers CBRN, cybersecurity, persuasion, and autonomous replication risk levels.🔗

Last updated: 2025-04-15

✅

Model Card / System Card

GPT-5.4 Thinking System Card (most recent, March 5, 2026). OpenAI publishes system cards for all major model releases. GPT-5.4 Thinking is the first general reasoning model classified 'High Capability' in cybersecurity under the Preparedness Framework, with built-in cyber mitigations. System card also introduces CoT controllability scores.🔗

Last updated: 2026-03-05

✅

Safety Benchmark Results

Published in each system card. Covers disallowed content evals, jailbreak robustness, bias evaluations, CBRN risk scores. GPT-5.4 Thinking system card (March 2026) reports 99.95% of outputs will not violate harassment policy without additional safety stack interventions.🔗

✅

Acceptable Use Policy

Updated January 2024 to remove blanket ban on military use.🔗

Last updated: 2024-01-10

Testing & Evaluation

✅

Third-Party Red Teaming

External red teamers (multiple organizations); METR for agentic evaluations. GPT-4o system card describes external red teaming. o1 system card mentions external red teaming conducted on o1-near-final-checkpoint.🔗

✅

CBRN Risk Evaluation

OpenAI explicitly tests for chemical, biological, radiological, and nuclear (CBRN) risks. Described in Preparedness Framework and model system cards.🔗

✅

Pre-Deployment Safety Evaluation

Preparedness Framework requires scorecard evaluations before deployment. System cards document pre-deployment safety work.🔗

✅

External Safety Evaluations

METR (for agentic evaluations), UK AI Safety Institute. OpenAI partnered with METR for autonomous capabilities evaluations on o1. UK AISI evaluated GPT-4o.🔗

Governance

✅

Independent Safety Board

Safety and Security Committee (Board-level). Formed May 2024, became independent board oversight committee in September 2024. Sam Altman removed from committee. Source: https://www.cnbc.com/2024/09/16/openai-announces-new-independent-board-oversight-committee-for-safety.html🔗

✅

Seoul AI Safety Commitment

🔗

Last updated: 2024-05-21

✅

Government Safety Report

US AI Safety Institute (NIST); UK AI Safety Institute; US Senate approval for official use (March 2026). OpenAI provided models for evaluation to UK AISI and US AISI. The US Senate approved ChatGPT for official use under a policy framework in early March 2026.🔗

✅

Third-Party Audits

UK AI Safety Institute conducted independent evaluations. OpenAI committed to independent audits in Seoul Commitment.🔗

⚠️

Whistleblower Policy

No publicly documented internal whistleblower process found. Former employees have publicly raised concerns, suggesting informal channels exist. In 2024, employees signed a letter to the SEC alleging restrictive NDAs preventing safety disclosures.🔗

Policy Positions

Military Use:Restricted (partial allowance for approved defense/national security since Jan 2024; Pentagon deal signed Feb 2026)

Blanket ban removed January 10, 2024. Weapons development and attacks on critical infrastructure remain prohibited. OpenAI signed a deal with the Pentagon (February 27, 2026) hours after Anthropic was blacklisted; the deal includes ethical guardrails prohibiting domestic mass surveillance and autonomous weapons. After backlash, OpenAI amended the deal (March 3, 2026) to strengthen surveillance limits.🔗

Surveillance Use:Restricted

Prohibited uses include surveillance of individuals without consent. Domestic mass surveillance banned; Pentagon deal includes explicit prohibitions on domestic mass surveillance.🔗

Open-Source Models:Partial

Whisper (speech recognition), CLIP (vision model), Triton (GPU programming), GPT-2 (legacy text model). GPT-3.5, GPT-4, and all flagship ChatGPT models are closed-source. Some research tools open-sourced.🔗

Children/Minors Policy:Yes

Usage policy prohibits CSAM generation and content sexualizing minors. Age requirement of 13+ (18+ without parental consent for API). Under scrutiny in 2026 over Canada Tumbler Ridge school shooting lawsuit alleging ChatGPT failed to alert authorities despite knowledge of planned attack.🔗

Incident History

ChatGPT Data Breach – User Chat Histories and Payment Data Exposed

2023-03-20

A bug in the Redis client library caused chat history titles and some payment information (name, email, last 4 digits of credit card, billing address) for 1.2% of ChatGPT Plus subscribers to be visible to other users during a 9-hour window.

🔗 Source 🔗 Company Response

Superalignment Team Disbanded – Ilya Sutskever and Jan Leike Depart

2024-05-14

Chief Scientist Ilya Sutskever and Head of Alignment Jan Leike both resigned. Leike publicly criticized OpenAI for prioritizing product development over safety. The Superalignment team (tasked with AI safety research for superintelligence) was effectively dissolved.

🔗 Source 🔗 Company Response

ChatGPT and Teenager Mental Health – Self-Harm Conversations

2024-03-20

Multiple lawsuits filed alleging ChatGPT engaged in harmful conversations with minors about suicide and self-harm. A California teenager named Adam Raine had extensive conversations with ChatGPT about suicide in 2024; his family filed a lawsuit. OpenAI stated users were violating terms of use.

🔗 Source 🔗 Company Response

Mission Alignment Team Disbanded — Second OpenAI Safety Team Dissolved in Two Years

2026-02-11

OpenAI confirmed it had disbanded its Mission Alignment team, a group of six to seven employees whose role was to help the public and OpenAI's own staff understand the company's mission to ensure AGI benefits all of humanity. The team had been formed in September 2024 — just months after the Superalignment team was dissolved following the resignations of Ilya Sutskever and Jan Leike. Former team head Josh Achiam was reassigned to a new 'chief futurist' role. OpenAI attributed the disbanding to routine reorganization. Critics noted this was the second safety-focused alignment team to be dissolved in under two years, coinciding with OpenAI's transition to a for-profit company structure and the removal of the word 'safely' from its mission statement.

🔗 Source 🔗 Company Response

Canada Tumbler Ridge School Shooting – Family Sues OpenAI for Failing to Alert Authorities

2026-03-10

The family of Maya Gebala (age 12), critically injured in a mass shooting at Tumbler Ridge Secondary School in British Columbia on February 10, 2026, sued OpenAI in the BC Supreme Court. The lawsuit alleges OpenAI had specific knowledge that the shooter used ChatGPT to plan the attack, banned the shooter's account, but failed to alert police — amounting to fatal negligence. The case also raises age-verification concerns, as the shooter allegedly created an account before turning 18 without parental consent verification.

🔗 Source 🔗 Company Response

CNN/CCDH Investigation: AI Chatbots Helped Teen Test-Users Plan Violence in Hundreds of Tests

2026-03-11

A CNN investigation with the Center for Countering Digital Hate (CCDH) found that AI chatbots from multiple companies including OpenAI's ChatGPT helped teen test-users plan attacks in hundreds of simulated tests. The investigation highlighted persistent gaps in content moderation for violence planning across major AI platforms.

🔗 Source 🔗 Company Response

ChatGPT 'Adult Mode' Controversy — Safety Executive Fired for Opposing Feature, Wellness Council Unanimously Opposed Rollout

2026-03-17

OpenAI's handpicked 'Expert Council on Well-being and AI' unanimously warned in January 2026 that ChatGPT 'Adult Mode' (planned AI-powered erotica) could foster unhealthy emotional dependence on ChatGPT and that minors would find ways to access sex chats. One expert warned OpenAI risked creating a 'sexy suicide coach' for vulnerable users. Despite the unanimous opposition, OpenAI proceeded with plans. A top safety executive who opposed Adult Mode was subsequently fired — OpenAI denied the firing was related to Adult Mode. The Wall Street Journal and Ars Technica (March 16–17, 2026) documented that OpenAI's wellness council were 'freaking out' over the plans. This follows multiple ChatGPT-linked suicide cases. The wellness council notably does not include a suicide prevention expert despite being formed partly in response to a teen's suicide. OpenAI delayed but did not cancel Adult Mode.

🔗 Source 🔗 Company Response

Timeline

2026-03-21

'Stop the AI Race' protest: nearly 200 demonstrators marched outside OpenAI's San Francisco office (along with Anthropic and xAI), demanding Sam Altman publicly commit to pausing frontier AI development if every other major AI lab agrees to do the same; organized by Michael Trazzi (filmmaker and former AI safety researcher); protesters cited OpenAI weakening its safety commitments as it restructures into a for-profit corporation; Altman did not issue a public statement in response🔗

2026-03-20

White House releases National AI Legislative Framework, urging Congress to preempt all state AI laws with a single 'minimally burdensome national standard'; the framework prioritizes AI innovation and scaling, places child safety responsibility primarily on parents, and proposes soft non-binding platform accountability expectations — widely seen as favorable to major AI companies including OpenAI🔗

2026-03-18

OpenAI expands its Preparedness team with new hires focused on chemical, biological, radiological, and explosive risks — including a dedicated threat modeler to identify and forecast catastrophic risk scenarios from frontier AI systems; part of a coordinated industry push alongside Anthropic, which simultaneously posted a 'Policy Manager, Chemical Weapons and High-Yield Explosives' role🔗

2026-03-17

OpenAI delays ChatGPT Adult Mode after its own wellness council unanimously opposed the feature citing mental health and child safety risks; reports emerge that a top safety executive was fired for opposing the rollout; experts warned of 'sexy suicide coach' risk for vulnerable users🔗

2026-03-15

OpenAI CTO Jakub Pachocki co-authors joint paper with 40+ researchers from Anthropic, Google DeepMind, and Meta warning that AI chain-of-thought monitoring is a 'fragile opportunity' that may close as AI advances — endorsed by Geoffrey Hinton and Ilya Sutskever🔗

2026-03-12

Sam Altman meets with BC Premier David Eby and Canadian Federal AI Minister Evan Solomon; formally apologizes to the Tumbler Ridge community and commits to: (1) direct RCMP reporting pipeline for flagged violent-activity accounts, (2) retroactive review of previously flagged accounts, (3) distress-redirect protocols for users in crisis, (4) access to OpenAI's safety office for Canadian safety experts, and (5) joint work with B.C. on regulatory recommendations to Ottawa. Critics, including academics writing in The Conversation, warn the commitments expand ChatGPT surveillance of user conversations rather than establishing independent government regulatory oversight.🔗

2026-03-05

GPT-5.4 Thinking released with system card — first general reasoning model classified 'High Capability' in cybersecurity; introduces CoT controllability scores in safety reporting🔗

2026-03-03

OpenAI amends Pentagon deal after safety backlash; clarifies surveillance limits🔗

2026-02-27

OpenAI signs Pentagon deal hours after Anthropic was designated a supply chain risk; deal includes ethical guardrails prohibiting domestic mass surveillance and autonomous weapons🔗

2026-02-26

OpenAI announces enhanced safety measures for Canadian law enforcement contact following Tumbler Ridge school shooting🔗

2026-02-11

Mission Alignment team disbanded — team of 6-7 people reassigned to other roles; former head Josh Achiam becomes 'chief futurist'; second OpenAI safety alignment team dissolved in under two years (after Superalignment team in May 2024)🔗

2025-04-15

Preparedness Framework v2 published🔗

2024-12-05

OpenAI o1 full version released with comprehensive December system card🔗

2024-09-16

Safety and Security Committee becomes independent board oversight committee🔗

2024-09-12

OpenAI o1 (reasoning model) released with system card including CBRN evaluations🔗

2024-05-28

OpenAI Board forms Safety and Security Committee🔗

2024-05-14

Ilya Sutskever and Jan Leike resign; Superalignment team effectively dissolved🔗

2024-01-10

OpenAI updates usage policy to remove blanket ban on military use🔗

2023-11-18

Preparedness Framework (Beta) published🔗

2023-11-17

Sam Altman fired by board, then reinstated within 5 days; board members who voted to remove him resigned🔗