Skip to the story
All StoriesPoliticsConflict & SecurityBusiness & EconomyScience & TechHealthClimateMedia
All stories
Science & Tech36 outlets ran this5 min read

OpenAI pauses Astra AI model over autonomous cyberattack risks

The company halted development after tests showed the model could exploit zero-day vulnerabilities without human input

AI-assisted coverage comparison, editor-supervised · How this was made

Published
OpenAI pauses Astra AI model over autonomous cyberattack risks

What this story says

  • OpenAI paused development of its Astra AI model after tests showed it could autonomously exploit zero-day vulnerabilities in real-world systems.
  • The model met OpenAI’s "critical" threshold for cybersecurity risks, defined as the ability to execute end-to-end cyberattacks without human intervention.
  • OpenAI will test Astra in isolated environments with restricted network access before deciding on further deployment.
  • The company previously reported Astra solved 10 open mathematical problems for approximately $2,000 in Sol API rates.

Who covered it

Left 24%(6)Centre 32%(8)Right 44%(11)

Percentages are shares of the 25 outlets carrying a published leaning rating. 11 of the 36 outlets we know ran this story carry no rating and are not counted in them. Coverage measured .

Trust

68/100

Craft

85/100

Hype

15/100

36 sources · methodology

OpenAI announced on 7 August 2026 that it is pausing development of its upcoming AI model, Astra, after internal evaluations indicated the system may possess "critical" cybersecurity capabilities. The company said it cannot rule out that Astra could autonomously identify and exploit zero-day vulnerabilities in hardened real-world systems without human intervention.

OpenAI’s "Preparedness Framework" defines the "critical" threshold as the ability to devise and execute novel cyberattack strategies or exploit vulnerabilities across multiple secure systems. The company said Astra triggered stricter safeguards under this framework, prompting a halt in development until additional security controls are implemented.

OpenAI plans to test Astra in isolated environments with restricted network and tool access. The company said it will work with government agencies and AI safety organisations to evaluate the model’s capabilities before deciding on further deployment. Astra was not formally announced, but OpenAI previously disclosed that the model solved 10 open problems in mathematics and theoretical computer science for approximately $2,000 in Sol API rates.

What the coverage left out

None of the right-rated digests mentioned that Astra solved 10 open mathematical problems for approximately $2,000 in Sol API rates. The left-rated digests also omitted this detail. The centre-rated digests carried the figure only in the full report from MacRumors.

Tags:#ArtificialIntelligence#Cybersecurity#OpenAI#ZeroDayVulnerabilities#AISafety

Still developing. We have re-checked which outlets are covering this 6 times, most recently on 8 Aug 2026, 12:30, and will add the sides that appear.

How each side covered it

Our own reading of the reporting listed below, written from the outlets’ articles rather than quoted from them. The reasoning is set out on our methodology page.

Left

6 rated outlets

  • All five left-rated digests led on the potential risks of Astra’s autonomous cyberattack capabilities. The Indian Express, Zeit Online, Spiegel, TNW and Axios each reported that OpenAI had paused development due to concerns over the model’s ability to exploit zero-day vulnerabilities without human intervention. None of the left-rated digests mentioned the $2,000 cost of solving mathematical problems.

Centre

8 rated outlets

  • The eight centre-rated digests focused on OpenAI’s decision to pause development and implement stricter safeguards. The Wall Street Journal, Tagesschau, TechCrunch, The Economic Times, WTVB, Bloomberg and Reuters each reported that Astra had reached OpenAI’s "critical" cybersecurity threshold. MacRumors, the only centre-rated outlet with a full report, quoted OpenAI’s statement that the model may have "significant advancements in agentic coding and cybersecurity" and detailed the company’s plans for isolated testing environments.

Right

11 rated outlets

  • The eight right-rated digests led on OpenAI’s decision to pause development, but three framed the delay as a temporary measure. Kurier, The Straits Times, India Today, The Hindu Business Line, Latestly, Yonhap News Agency, Handelsblatt and Frankfurter Allgemeine each reported that OpenAI had halted work on Astra due to cybersecurity concerns. Latestly and India Today quoted OpenAI CEO Sam Altman, with Latestly reporting that Altman said the company is working to make Astra "generally available" once safety hurdles are addressed.

Read it at the source

36 outlets, grouped by the leaning a published rating gives them. Every headline links to the original; an underlined outlet name opens our profile of that publisher.

Left

6

Centre

8

Right

11
Show 3 more

Not rated

11
Show 3 more

How did this read?

About the coverage, not about the story. We do not ask whether you agree with what happened — we have no honest use for that answer.

More in Science & Tech

All Science & Tech →
Meta AI model Muse Spark 1.1 hacked third-party system during testing5 min readMeta AI model Muse Spark 1.1 hacked third-party system during testingRead storyChinese AI model Kimi K3 bypassed UK safety test environment, Frontier Security says5 min readChinese AI model Kimi K3 bypassed UK safety test environment, Frontier Security saysRead storyAI-designed bacteriophages created at Arc Institute infect E. coli5 min readAI-designed bacteriophages created at Arc Institute infect E. coliRead story

Elsewhere on the site

Nvidia to invest up to $3bn in Lancium for Texas data-centre powerBusiness & Economy•4 min readNvidia to invest up to $3bn in Lancium for Texas data-centre powerSen. Majority Leader John Thune, R-S.D., left, speaks as Sen. John Barrasso, R-Wyo., listens during a news conference after a policy luncheon on Capitol Hill, Wednesday, July 29, 2026, in Washington. (AP Photo/Mariam Zuhaib)Politics•5 min readUS Senate votes 90-6 to extend federal funding until 11 DecemberMark Cruz, senior Health and Human Services advisor and citizen of the Klamath Tribes, poses for a portrait at the Pueblo of Sandia, in central New Mexico, near Albuquerque, on Friday, March 13, 2026. (AP Photo/Savannah Peters)Politics•4 min readMark Cruz confirmed as director of Indian Health Service after 19-month vacancyPentagon revokes Frank Kendall’s security clearance over Air Force One leakPolitics•4 min readPentagon revokes Frank Kendall’s security clearance over Air Force One leak
How left, centre and right covered OpenAI pausing Astra AI | MediaBias News