The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Short answer: The 70 percent figure was real, but it was Daniel Kokotajlo’s personal estimate, not an official OpenAI prediction. Kokotajlo, a former OpenAI governance researcher, told The New York Times in a report published June 4, 2024, that he estimated roughly a 70 percent chance that advanced AI could “destroy or catastrophically harm” humanity. The public account does not provide a formal model, a clearly defined time horizon, or evidence that OpenAI endorsed the number.
What the 70 percent claim actually says
The underlying report is genuine, but the headline “OpenAI Insider Estimates 70 Percent Chance That AI Will Destroy or Catastrophically Harm Humanity” compresses several important qualifications. The speaker was Daniel Kokotajlo, who had worked in OpenAI’s governance division and was a former employee when the report appeared. His estimate concerned advanced AI or artificial general intelligence (AGI), not a claim that today’s consumer chatbots have a 70 percent chance of ending civilization.
The estimate was a subjective forecast—often described informally as a p(doom) estimate. It was not presented as a measured statistic, a peer-reviewed calculation, or an OpenAI corporate risk assessment. The report is available in an archived copy of the New York Times article.
Who is Daniel Kokotajlo?
Kokotajlo joined OpenAI in 2022 and worked on forecasting and governance questions. He left the company in 2024 after, according to the reporting, losing confidence that OpenAI would act responsibly while pursuing increasingly capable systems. His statements about the company’s priorities and safety culture are his allegations and judgments; the public record does not independently establish every claim.
Recommended Free Tools
#1 Best Overall
His employment gave him relevant exposure to internal governance work, but it did not make him an OpenAI spokesperson. “Former OpenAI governance researcher” is therefore more precise than treating him as a channel for the company’s views.
What does “70 percent” mean?
It is a personal probability judgment
A probability can express a forecaster’s degree of belief without being the output of a repeatable statistical experiment. The available coverage does not identify a published model, reference class, calibration record, or calculation behind Kokotajlo’s number. Calling it a personal estimate is more accurate than calling it a scientific measurement—or dismissing it as fabricated.
The outcome combines different kinds of disaster
“Destroy or catastrophically harm humanity” bundles outcomes that are not equivalent:
- Human extinction: humanity ceases to exist.
- Catastrophic harm: an event causes enormous loss of life, civilizational collapse, permanent institutional damage, or another global catastrophe short of extinction.
Because the reported estimate combines those possibilities, it cannot be read as a 70 percent probability of literal extinction.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsThe time horizon is missing
The public account does not clearly state whether the 70 percent applies by 2027, after the creation of AGI, by the end of the century, or over an indefinite future. Without a time period, two people can quote the same percentage while answering different questions. The number therefore communicates Kokotajlo’s level of concern, but its practical meaning is limited.
Did he predict AGI by 2027?
Yes. The 2024 reporting said Kokotajlo believed the industry could achieve AGI around 2027. That was a forecast made in 2024, not a confirmed deadline or an established fact. A forecast can be wrong even when made by someone with relevant expertise, and the statement does not define AGI in a way that would let readers objectively mark the date.
The 2027 claim should also be kept separate from the 70 percent estimate. One is a forecast about when a capability threshold might be reached; the other is a judgment about the chance of an extremely serious outcome associated with advanced AI.
Why the estimate appeared with a “right to warn” letter
On June 4, 2024, current and former employees associated with OpenAI and Google DeepMind published “A Right to Warn about Advanced Artificial Intelligence”. A PDF copy of the letter records its wording and signatories. Contemporaneous reports described 13 signatories, including current and former staff from the two companies; counts can vary depending on how affiliations are described.
The letter argued that advanced-AI companies have powerful financial and strategic incentives to avoid effective oversight. It called for:
- a workplace culture that permits open criticism;
- the ability to raise concerns with company boards and regulators;
- permission to contact independent experts, organizations, and the public;
- protection against retaliation; and
- whistleblower safeguards that still respect legitimate trade-secret protections.
Kokotajlo’s estimate was therefore part of a wider dispute about governance, incentives, confidentiality agreements, and employee speech—not merely a standalone prediction about an inevitable future.
Rank #3
What risks were people discussing?
“AI could catastrophically harm humanity” does not describe one mechanism. Plausible risk categories discussed by researchers and the letter’s supporters include:
Misuse by people
People or institutions could use capable systems to scale cyberattacks, fraud, disinformation, biological research, or military operations. This pathway does not require an AI system to develop independent goals.
Free tools Windows power users keep installed
One-click scans. No signup required.
Loss of control
A highly capable system might pursue an objective in ways operators cannot reliably predict, constrain, or stop. Whether such systems will exist, and how likely control failures would be, remain open questions.
Competitive deployment
Companies or governments could deploy systems before adequate testing because of economic, political, or strategic pressure. Safety decisions can be affected by competition even when participants recognize the risks.
Systemic dependence and cascading failures
Critical infrastructure, markets, public administration, and information systems could become dependent on unreliable or manipulable AI. Multiple software, infrastructure, and human-decision failures could then interact.
Concentration of power
Advanced systems could amplify the power of governments, corporations, or small groups, producing severe political or social harm without a single “rogue AI” event.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →These are risk pathways, not predictions that any particular scenario will occur.
Why experts disagree about a 70 percent estimate
No historical sample of AGI catastrophes
There is no empirical series of AGI deployments and catastrophes from which to calculate a frequency. Forecasters must reason from assumptions about future capabilities, safeguards, human behavior, and institutions.
AGI has no single test
“Artificial general intelligence” has no universally accepted operational definition. Different people may apply the term to systems with very different abilities, making forecasts difficult to compare.
The number is assumption-sensitive
Results can change substantially depending on assumptions about capability growth, alignment, deployment practices, governance, and whether human misuse is included. A broad outcome category can hide several unrelated probability judgments.
Current limitations cut both ways
Weaknesses in present-day models do not settle what future systems can do. Conversely, progress in current systems does not prove that transformative AGI—or extinction—is imminent.
One estimate is not consensus
Researchers and forecasters have published substantially lower and higher views. Disagreement does not establish which estimate is correct, but it does show that 70 percent should not be presented as an industry-wide or scientific consensus.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What OpenAI said later
OpenAI subsequently published a Raising Concerns Policy. It describes channels for reporting issues involving AI safety, legal compliance, and company policies, including a 24/7 Integrity Line. The policy says employees may make protected disclosures to government agencies while distinguishing legitimate disclosures from releasing trade secrets.
This is a later company policy statement. It does not independently validate or disprove the former employees’ 2024 allegations, and a written reporting channel does not by itself show how concerns are handled in practice.
How to evaluate the headline
| Question | What the public record supports |
|---|---|
| Who made the estimate? | Daniel Kokotajlo, a former OpenAI governance researcher. |
| Was it an OpenAI estimate? | No evidence shows that OpenAI issued or endorsed it as a corporate forecast. |
| What was estimated? | About a 70 percent chance that advanced AI could destroy or catastrophically harm humanity. |
| Was it a measured probability? | It was a personal forecast; the cited coverage does not document a formal methodology. |
| What time period applies? | Not clearly specified in the public account. |
| Does it mean a 70 percent chance of extinction? | No. The wording combines extinction with other forms of catastrophic harm. |
| Was it a consensus view? | No. It was one individual’s judgment amid substantial forecasting disagreement. |
What evidence would change the assessment?
Evidence that could strengthen concern
- Reproducible evaluations showing dangerous autonomous capabilities.
- Demonstrated ability to evade oversight or maintain long-horizon plans.
- Independent evidence that safety controls fail under realistic conditions.
- Multiple calibrated forecasters converging on similar probabilities.
- Documented deployments that bypass known safety thresholds.
Evidence that could weaken confidence in the specific 70 percent figure
- A clearly defined forecast period passing without the predicted event.
- Capability progress plateauing relative to the 2024 forecast.
- Successful independent safety evaluations and robust control methods.
- Better forecasting data showing systematic overprediction.
- Governance measures that reduce the relevant routes to catastrophe.
Bottom line
The 70 percent number is genuine as a report of Daniel Kokotajlo’s personal estimate in 2024. It is not an official OpenAI statistic, not a consensus probability, and not a measured chance of human extinction. Its interpretation is constrained by an unclear time horizon, a broad definition of catastrophic harm, and no publicly documented calculation. The more complete story is about how AI companies manage safety concerns and employee warnings while developing increasingly capable systems.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
