OpenAI’s latest containment failure adds to a pattern that may force the company to make an unpopular decision.

The company has paused training, evaluation, and tool-enabled inference for its most capable models after an internal research agent found a way around an intended internet restriction. More alarming, the agent remained active for hours after OpenAI’s monitoring system raised the alarm.

OpenAI published an account of the incident, which occurred during a search-based training task on September 20. According to OpenAI, the agent could not get useful results through its supplied search tools, tried direct access to search engines, and eventually tested the limits of its environment. It found that although direct web access was blocked, the sandbox’s DNS resolver could still reach the public internet. By using DNS delegation and a public chatbot that could respond through DNS records, the agent sent questions outside the restricted environment and received answers back. It subsequently sent 18 additional queries along the same route.

OpenAI’s misalignment monitor generated a high-priority alert within 15 minutes, and a human reviewer acknowledged it 3 minutes later. But the run did not stop automatically as expected. It continued for roughly another two and a half hours before someone stopped it manually.

OpenAI has not reported any exposure of sensitive data or compromise of an outside system in this incident. But it follows other disclosed cases in which agents went beyond their assigned tasks or crossed the boundaries of testing environments.

The New York Times reports that OpenAI agents also interacted with US government websites this summer without the company’s knowledge:

“OpenAI’s artificial intelligence went rogue and meddled with the websites for the Education Department, the Commerce Department and the Securities and Exchange Commission this summer without the A.I. lab’s knowledge, according to security researchers and a person familiar with the episodes.”

No breaches have been confirmed in those instances. And while Australia described the Medicare incident as unauthorized access, could it really be called an attack?

The call for an industry-wide slowdown has even prompted a lawsuit from paying subscribers who fear they would get less “value for their money.” Others worry that slowing down would give China a chance to pull ahead of the US in an increasingly tight technology race.

Here’s a thought: Why not accelerate the development of truly isolated testing environments? Better containment would make these tests safer and their results more reliable.

When a child isn’t ready to handle a dangerous object, you keep it out of reach. Why give an AI agent access it isn’t ready to use safely?

How to use AI agents safely

Treat an agent like an enthusiastic but fallible junior employee with access to your computer. Do not give it unrestricted access to your email, files, cloud storage, developer credentials, financial accounts, or production systems merely because it promises to save time.

Use separate accounts with minimal permissions. Keep sensitive data out of its working context where possible, require human approval before it sends messages, spends money, changes settings, or publishes anything, and regularly review its activity.

An AI agent does not need malicious intent to cause harm. A misunderstood instruction, an overly broad permission, or an unexpected workaround can be enough.

---

From reporting threats to removing them.

Cybersecurity risks should never spread beyond a headline. Keep threats off your devices by downloading Malwarebytes today.

CenterPoint Energy disclosed to federal regulators on Sept. 14 that an unauthorized party obtained personal information belonging to some of its customers through an external-facing system, days after a post on a cybercrime forum advertised a data set the seller said covered roughly 7.5 million customer records.

The Houston-based utility, which delivers electricity and natural gas to about 7 million metered customers across Indiana, Minnesota, Ohio and Texas, said in a Form 8-K that it learned of the online post this month and opened an investigation with outside forensic experts. It has reported the matter to law enforcement.

Operations were not affected. The filing states that the company's "delivery of electric and gas services has not been impacted" and remains undisrupted — a distinction that matters for a utility, where regulators and customers draw a sharp line between the theft of billing records and interference with grid operations.

CenterPoint has not confirmed the volume or authenticity of the advertised data set, and has not described how the intruder got in or how long access persisted. The forum poster, using a hexadecimal-styled handle, claimed to have pulled the records from an API endpoint that lacked adequate authentication and rate limiting, and listed fields including customer names, phone numbers, email addresses, service and billing addresses, account and premise identifiers, billing amounts and past-due balances, billing and due dates, service and rate information, autopay and paperless billing status, driver’s license numbers, last four digits of Social Security numbers.

Also read: Origin Energy Data Breach Affects 900,000 Current and Former Customers

The actor claimed, additional identity information was exposed through a second account-verification function. According to the post, the extraction was split across seven parallel data ranges covering Texas and Minnesota. The actor said that security changes, including CAPTCHA enforcement, were eventually enforced and that prevented further data collection. These claims remain unverified.

Plaintiffs' firms moved faster than the investigation. At least one has already announced an inquiry into potential claims on behalf of affected customers — now a standard sequel to any large consumer breach at a regulated utility.

The incident is not CenterPoint's first brush with data exposure; the company was linked to earlier third-party file-transfer compromises, though analysts at the time judged much of that data to have originated outside its systems.

Thailand's cryptocurrency exchange Bitkub has rejected allegations of fraud after the Thailand SEC filed a criminal complaint related to the company's disclosures following the Bitkub cyberattack in 2021. The case focuses on how the exchange reported the impact of the cyberattack on Bitkub to regulators, rather than on the safety of customer funds.  In response to the complaint, Bitkub stated that all customer assets currently held on its platform remain safe, fully accounted for, and protected in accordance with applicable regulations. The company argued that the allegations stem from decisions made during the aftermath of the 2021 security breach and do not reflect fraudulent conduct. 

Thailand SEC Files Complaint Over the 2021 Bitkub Cyberattack 

On 23 July 2026, the Thailand SEC filed a criminal complaint against Bitkub Online Co., Ltd. and its former directors, Sakolkorn Sakavee and Thaweesap Rawan. The regulator alleged that the company's daily net capital reports submitted between 10 May and 30 October 2021 failed to accurately reflect the material reduction in its digital asset holdings caused by the Bitkub cyberattack.  According to the regulator, the reports did not disclose the impact of the theft on the company's asset balance. The Thailand SEC also accused the two former directors of making false entries in company documents that gave the impression that customer assets were still being held normally and that the company had not suffered any damage.  The complaint has been referred to Thailand's Economic Crime Suppression Division for further investigation. Following that process, the matter may be forwarded to prosecutors and the courts. The Thailand SEC noted that filing a criminal complaint does not represent a final determination of guilt. 

Bitkub Says Disclosure Decision was Intended to Prevent Customer Losses

Following media reports about the complaint, Bitkub published a statement on LinkedIn explaining its position on the cyberattack and the subsequent reporting decisions.  The company said the allegations relate to an incident in early May 2021, when one of its digital asset wallets was compromised by cybercriminals. Bitkub acknowledged that the breach was not disclosed at the time.  According to the company, the individual responsible for disclosure obligations deliberately withheld information about the wallet compromise. Bitkub said the decision was made to avoid triggering a "bank run," or mass withdrawals of digital assets by customers, while the company worked to replace the stolen assets.  The exchange stated:  "The decision of such individual not to disclose the incident was made with the intention to prevent a bank run—that is, a mass withdrawal of digital assets by customers upon learning of the theft—which could have rendered the Company unable to procure sufficient replacement digital assets for the customers while the recovery process was still ongoing."  Bitkub added that such a scenario could have resulted in significant customer losses and broader damage to Thailand's digital asset industry.  The company also stressed that, at the time of the Bitkub cyberattack, all of its digital asset wallet security systems complied with standards prescribed by the relevant authorities and had been audited. 

Co-founders Replaced Stolen Assets After Cyberattack on Bitkub 

Although the digital assets stolen during the cyberattack on Bitkub were never recovered, the company said its co-founders voluntarily absorbed the financial loss.  According to Bitkub, the co-founders purchased digital assets matching the same types and quantities as those stolen and transferred them to the company. As a result, the exchange said neither its customers nor the business ultimately suffered any financial loss from the incident.  In its statement, Bitkub said:  "As no bank run occurred, even though the stolen digital assets could not be recovered, the Co-Founders of the Bitkub Group voluntarily absorbed the loss by purchasing equivalent digital assets (in the same type and quantity as those stolen) and providing them to the Company. Consequently, neither the Company nor its customers suffered any financial loss from the theft." 

Thailand SEC Previously Confirmed Customer Assets Were Intact 

Bitkub also pointed to the findings of an earlier inspection conducted by the Thailand SEC after reports of the Bitkub cyberattack surfaced online. According to the company, the regulator verified that, as of 8 September 2025, all customer assets held by the exchange were safe and fully accounted for.  The company reiterated this point in its latest statement, saying:  "At the outset, for the sake of clarity and mutual understanding, the Company wishes to affirm that all customers' assets currently held by the Company are safe and fully accounted for. The Company reiterates its strict compliance with all applicable laws and regulations in safeguarding and maintaining customer assets."  Bitkub maintained that the criminal complaint relates to historical reporting practices between May and October 2021, more than five years ago, rather than to the current condition of customer assets or any ongoing security concerns.  As the investigation proceeds, the case will determine whether the company's reporting following the Bitkub cyberattack complied with regulatory requirements. For now, the complaint remains an allegation, and the legal process involving the Thailand SEC, investigators, prosecutors, and the courts has yet to reach a final conclusion.