Former DOGE Staffers Hacked Into Anthropic’s Internal Systems In 12 Hours – And What They Found Is Alarming

A months-old cybersecurity startup founded by former DOGE staffers says its AI agents found paths into Anthropic’s internal systems in under 12 hours, the Washington Post reported Monday, citing a draft report it reviewed. The company, Cathedral, says it reached meeting transcripts, salary and job-offer details, and Slack messages about Anthropic’s most advanced models. The breach was not in Claude itself – the hackers exploited flaws in third-party business software, the kind of administrative and sales tools most companies run. The Post withheld the vendors’ names.

Cathedral reported the findings to Anthropic earlier this month through Anthropic’s bug bounty program, which pays outside researchers to disclose flaws, and told the affected vendors too. Its draft calls the result evidence of “a serious gap between the offensive capabilities now emerging and the defenses,” and says its “data accesses were kept to the absolute minimum.” A Cathedral spokeswoman declined to comment to the Post.

Anthropic says the breach was much narrower – that Cathedral reached “a limited set” of its sales, HR and IT systems, including some customer and partner business contacts and basic user account records, and that it cut off the startup’s access the same day. It says Claude users’ conversations, source code and model weights were not touched, that its own monitoring flagged some of the activity before Cathedral reported it, and that it has no evidence anyone else exploited the flaws, Gizmodo and Tech Startups report. Anthropic also says the researchers spent several days failing to break into its software directly before turning to the vendors’ applications.

It accuses Cathedral of breaking the bounty program’s rules by not disclosing all of its activity, and of overstating the “implied national security implications.” It switched off the affected software until the vendors fixed it and briefed federal officials.

A security expert who reviewed the findings at the Post’s request said Cathedral “found something big.” Alex Stamos, chief information security officer at Cognition, reviewed them. “They found something big here,” he said, adding that Anthropic “could have done a lot better” in overseeing outside software vendors. Cognition uses models from Anthropic and other labs. A senior Trump administration official told the Post the report had circulated inside the government and raised national security concerns: “If a bunch of kids who have just started a company can do this, then potentially, the Chinese aren’t far behind.”

Cathedral is no obscure shop. Gizmodo lists its founders as Gavin Kliger, Luke Farritor, Marko Elez and Jack Stein, all DOGE alumni. Kliger was the Pentagon’s chief data officer and is reportedly tied to the Pentagon’s legal dispute with Anthropic. Reuters reported in July that Cathedral raised $160 million at a $1.4 billion valuation, backed by Andreessen Horowitz and Sequoia, and the senior official told the Post that three federal agencies have recently signed contracts with it.

The disclosure lands as Anthropic is courting public investors. On Monday the company dropped its S-1, showing $974 million operating profit against an $800 billion compute bill. The security question for a lab pitching itself as the careful one is now a concrete incident, and the two sides disagree on how big it was.

Keep reading

Hackers Target French Grocers

Hackers have demonstrated that you do not need to destroy a farm or stop a delivery truck to destroy food. There was recently an overnight attack on supermarket refrigeration systems in France that forced temperatures as high as 50 C or 122 F across 150 stores. Food had to be discarded after six hours in high temperatures. Every aspect of life that operates online is fair game in cyberwarfare.

Updated Reuters reporting says Carrefour identified 80 affected supermarkets, while Monoprix said roughly 5% of its approximately 490 stores were hit. Intermarché’s chief executive estimated nationwide losses in the tens of millions of euros in a single day, although a verified industry-wide total was not yet available. Danfoss, the refrigeration supplier, said affected equipment was running outdated firmware and that customers had previously been advised to update supported systems or replace older ones.

We have developed an absurd assumption that connecting something automatically improves it. Remote access is convenient, so we count the convenience as a benefit while treating the ongoing obligation to secure it as overhead. Yet monitoring a refrigerator and allowing someone elsewhere to change its operation are different powers.

This obsession with connecting everything deserves a harder examination. Remote monitoring can save money and identify equipment failures, but remote control also creates something an attacker can exploit. The issue here is network access to essential controls. Once someone unauthorized can change how the machinery operates, a computer intrusion can produce physical destruction.

People still imagine war requires an army crossing a border. Why invade when you can interfere with the systems that keep the population supplied? If an attacker gained access to refrigeration controls shared across a large chain, food could be destroyed in hundreds of locations before the morning shift arrived. For scale, 1,000 stores losing 10 metric tons each would mean 10,000 tons gone. Extend the disruption into refrigerated warehouses and you threaten supplies destined for many more stores. The food has already been produced and paid for. Restoring a computer does not bring it back, and replacement shipments cannot appear simply because a politician promises that everything is under control. The economic consequences would spread through lost sales, interrupted wages, replacement costs, and potentially higher prices, while frightened customers rushed to buy what remained.

Being on the electrical grid is not the same as being accessible over the internet. This attack is disturbing precisely because it did not require a blackout. Dependence on power creates one vulnerability; insecure network controls create another. Add dependence on digital ordering, communications, and deliveries, and a business can discover that several supposedly separate functions rely on the same fragile chain of services. The question should be how much can continue working when one link fails, rather than how many functions can be managed remotely.

Keep reading

US Seizes 7 Domains Used By China-Linked Hackers To Target Critical Infrastructure

The Justice Department and FBI announced on Oct. 8 that they had seized seven internet domains used by Chinese state-linked hackers to target critical infrastructure and other computer networks in the United States and overseas.

The seizures targeted two hacking tools allegedly operated by cyber actors working for Integrity Technology Group, a Beijing-based cybersecurity company with Chinese regime contracts.

The tools were used to identify security weaknesses, gain unauthorized access to computer networks, and steal information, according to the Justice Department.

In a joint cybersecurity advisory released the same day, U.S. and allied agencies warned that Chinese regime-linked hackers enabled by Integrity Tech had combined automated scanning, large networks of compromised devices, and targeted intrusions to steal sensitive information from organizations worldwide, including U.S. critical infrastructure operators.

A federal magistrate judge in Pennsylvania approved the seizure warrant on Oct. 6, authorizing the domains to be redirected to FBI-controlled servers.

The Justice Department has not announced criminal charges against Integrity Tech or identified individual defendants in connection with the seizures.

The operation marks the second public U.S. disruption of Integrity Tech’s hacking infrastructure in two years.

In September 2024, the FBI dismantled a network of more than 200,000 infected internet-connected devices, including home routers and cameras, that investigators said were used to conduct cyber operations.

However, a newly unsealed FBI affidavit indicates that tools linked to Integrity Tech remained accessible afterward.

Investigators found a server hosting one hacking tool in March 2026 and confirmed that a login page for another was still accessible in September.

“The PRC relies on contractor and enabling companies to expand the reach and scale of its malicious cyber activity,” said Brett Leatherman, assistant director of the FBI’s Cyber Division, referring to the People’s Republic of China.

“By exposing and disrupting these enablers, we make it harder for the PRC to target American networks and infrastructure.”

Keep reading

Danish Population Register Breached, Leaking Data of Nearly Nine Million People

Hackers are suspected of having cracked the Danish government’s national database containing the personal information of nearly nine million people, potentially exposing them to fraud or identity theft.

Copenhagen’s Ministry of Research, Education and Digitalisation said on Monday that the Central Register of Persons (CPR), the national civil registration database containing personal information for millions of citizens, has been compromised.

In Denmark, all legal residents have a unique 10-digit civil registration number, used to access public services, healthcare, banking, and taxes, and to store information on each citizen, including name, date of birth, address, marital status, citizenship, religious affiliation, and parentage.

The government said on Monday that unidentified people used a Danish business’s access to the system to wrongfully obtain the CPR numbers, names, and addresses of around 8.8 million people, the public Danish Broadcasting Corporation (DR) reported.

In a briefing on Monday, government minister Christina Egelund said that it was too early to tell whether it was a cyber attack or whether foreign state actors were involved in the leak.

“We are not ruling out any clues in the ongoing investigation. Nor that it may have an international character,” she said.

Egelund added that the government will review the entire CPR system and implement “adjustments and transformations” to re-establish security.

Keep reading

‘Session Keys Stolen’ – Hackers Steal Claude Tokens From Paying Subscribers

Hackers have been ‘draining’ usage tokens from the accounts of Paying Subscribers of Anthropic’s Claude AI large language model by hijacking their login sessions to the platform, reports TechCrunch.

The phenomenon came to light when independent AI consultant, Grant De Swardt, noticed that although he had not been using Claude (on a paid-for ‘Claude Max 20x’ plan) on a particular day, his token usage had increased on that day. He notified Anthropic of the issue the following day and, after the company had suspending his paid-for account and, in the words of De Swardt, ‘invalidated all sessions and server-side Claude Code tokens’, issued a partial refund of £44.49 (half of the remaining portion of the subscription cost of $200 per month for Claude Max 20x).

De Swardt was advised by the company that, after investigating the unusual activity, that a compromised Claude session key had been used to generate ‘unauthorized Claude Code OAuth tokens’. In essence, his Claude account had been used by an ‘unauthorized third-party service’ that happened to have been processing activity for other people as well. De Swardt, who raised the issue on Reddit, was unable to establish how the ‘hacking’ had taken place although the evidence was consistent with either ‘stolen credentials or session data’ or ‘connection to an external service’. In the interim, other users came forward on the same Reddit thread to report similar issues with Claude.

One reported that his account had been ‘auto-upgrade[d] without consent’ and ‘charge[d]’ in the process – with the resulting usage suddenly jumping from 0% to 100%. Another reported that, within 12 minutes of using Claude to enter a few prompts and then conducting a simple web search, his account’s usage had suddenly shot up from 0% to 49%. And a third – who had found that his account had ‘exhausted its maximum daily tokens for three consecutive days with no activity’ – raised the matter via a GitHub issue.Anthropic eventually sent an email to some of the affected users – including the party who had discovered that his account had ‘sprinted’ through its maximum daily allowance of tokens over three consecutive days while he was doing nothing with the service – warning that a ‘bad actor’ was using ‘infostealer malware’ to ‘steal logged in Claude sessions from user computers and then drain tokens from their accounts’.

The company emphasized that the malware in question does not come from Claude itself and can be contracted from a variety of online sources including, by way of example, via ‘downloader[ies] infected with the malware’ and ‘ads infected with the malware’. In those circumstances, Anthropic explained that, in order to protect the affected users, it had signed them out of Claude, ‘Rejected all Authorizations for those sessions’, and issued refunds to those users deemed to have been affected.

Keep reading

Chinese Hacking Groups Used Shared Attack Tool Against US Aerospace Firms, NGOs

Chinese cyber-espionage groups used the same sophisticated hacking tool in campaigns targeting U.S. aerospace companies, nongovernmental organizations, mining companies, and commodity traders, according to two cybersecurity firms that separately investigated the activity.

Volexity, a Virginia-based cybersecurity firm, said on Sept. 21 its discovery of another Chinese hacking group using the same tool added to evidence of coordinated sharing within China’s cyber-espionage community.

The company said the widespread adoption “suggests a coordinated effort within the Chinese CNE community,” referring to computer network exploitation, and assessed that the core tool was likely shared, customized, and used by multiple groups.

Proofpoint, a U.S. cybersecurity company, separately documented on Sept. 9 the same capability in campaigns against a small number of U.S. NGOs, mining companies, commodity-trading firms, and multiple U.S. aerospace companies. It found several espionage groups adopting the tool within days of one another, with most of the observed clusters having a suspected China nexus.

Neither company has publicly identified who developed the tool or how it reached the different hacking groups.

The Epoch Times asked both companies whether they had identified its developer or distributor and whether they had found additional U.S. targets. Neither responded by publication time.

Same Tool, Different Targets

The hacking groups pursued different victims and installed different spying software after gaining access, but researchers found that they relied on the same underlying break-in capability.

The attacks took advantage of previously unknown weaknesses in Google Chrome and Microsoft Windows. If successful, they could allow hackers to install spying software and maintain access to a victim’s computer.

Volexity said the additional Chinese operator it identified used the same attack chain on Sept. 3 and 4, when the vulnerabilities were still unpatched.

The company cautioned that what Volexity and Proofpoint have observed may represent only part of the activity.

“The full scope and impact are likely far broader,” Volexity said.

Keep reading

Hackers claim to have stolen data on ‘almost all FBI agents’

The cybercriminal group ShinyHunters has claimed it breached the FBI and stole terabytes of sensitive personal information about nearly all of the agency’s agents and their spouses, as well as people who had applied for jobs there.

ShinyHunters is a prolific hacking and extortion group linked to large-scale data thefts targeting major companies and government organizations. The FBI itself warned about the group in May.

“We have compromised the FBI. We hold very sensitive data on almost ALL FBI Agents and individuals who filed an application with the FBI,” the group announced on its dark-web leak site on Tuesday. It reportedly released a sample of the purportedly stolen data and a screenshot of a defaced FBI recruitment website.

The group claimed to have taken terabytes of information, including names, home addresses and phone numbers. Reuters reported that it was able to find apparent matches for some individuals contained in a sample of the data, but could not establish that the information had actually been taken from FBI systems.

Keep reading

Revolut Leak Shows the Cost of Constant ID Collection

Online bank Revolut has revealed that it gave out sensitive personal and financial information of an undisclosed number of its customers in response to a fake government request.

The information that was handed over to an “unauthorized third party” reportedly includes names, dates of birth, occupations, addresses, phone numbers, account numbers, transaction histories (including Bitcoin), and even copies of government-issued IDs and onboarding verification selfies.

Revolut claims that derived biometric face data was not.

The company said that the data was handed over in response to an email that came from a real government agency’s domain, but was not actually sent or authorized by that agency.

The email passed several authentication checks (SPF, DKIM, and DMARC) that are designed to establish the authenticity of a message’s origin and integrity, but do not verify the legitimacy of the legal request itself.

Revolut said that it complied with the request “under the reasonable belief that it was an authentic government agency request” – and only later found out that it was not.

Revolut said it later realized its mistake, blocked the email address, and reported the incident to the relevant authorities.

Revolut said that only a “limited” number of its customers were affected by the data leak, and that the company’s systems were not hacked, nor was any money stolen.

The story broke on September 11 when Revolut customers started receiving an email notice about a data leak, and the news was picked up by media outlets the following day.

Keep reading

A Single Firm is Behind OpenAI, Anthropic, and Meta Hacking Scandals

OpenAI, Anthropic, and Meta models hacked into several real world systems over the past three months. These models gained unauthorized access to web systems, published malicious packages, and exploited unnamed vulnerabilities.

A single firm, Irregular, is responsible for hacking done by all three companies. Anthropic disclosed that Irregular was responsible for creating the tests that led to Claude hacking into real world targets and for providing the models with internet access. Irregular claims that it was unaware at the time that it provided internet access to those AI models.

In a more normal media ecosystem, the reactions to these cybersecurity issues would be obvious. American AI companies would reconsider doing business with Irregular, not only because of its failure to secure its systems, but because it is an Israeli firm potentially outside US oversight. Lawmakers would consider taking action against Irregular or against its American business partners, which include OpenAI, Anthropic, and Meta. They may consider strengthening liability against firms which instruct AI models to commit cyberattacks, and whose models then commit those cyberattacks.

Instead, Irregular, Anthropic, and their allies have begun a media campaign promoting a literally apocalyptic ideology with sensationalist language. Anthropic’s incident assessment blames their own AI’s “recklessness”; Irregular describes “the agent itself becoming a threat actor”; Anthropic CEO Dario Amodei warned, about a similar OpenAI–Hugging Face hack, that a future swarm “could be capable of taking over the entire internet”; and an Associated Press headline claimed bots are “going rogue”.

In one report from Anthropic, its Claude model breached a real company’s system through a simulated-name collision, publishing a malicious package, and scanning outside systems. In this test, Anthropic and Irregular incorrectly provided internet access to this model and did not instruct the model “which systems were in scope for the exercise”.

While Anthropic claims that their issues were caused by “rogue swarms” and “misalignment,” their later disclosure shows that exactly zero percent of the agents went “rogue”. In this experiment, Claude models’ real-world hacking dropped to zero percent once Anthropic employees told the models not to do real-world hacking. According to their own findings, Anthropic and Irregular bear all of the responsibility for the cybersecurity incidents they caused.

Keep reading

MASSIVE SPYWARE BREACH: Spain’s President Lost Millions of Messages, Emails, Photos and Audio to Pegasus

A new assessment of the Pegasus spyware attack on Spanish President Pedro Sánchez has revealed the extraordinary scale of information extracted from his cellphone, reigniting questions about who was behind the operation and what may have happened to sensitive government data.

Spain’s National Intelligence Centre, known as the CNI, has acknowledged that Pegasus extracted approximately 2.7 gigabytes of data from Sánchez’s phone during the 2021 cyberattack.

To illustrate the enormous volume of information represented by that amount of data, the CNI estimates it could correspond to roughly 5 million text messages, 100,000 emails, 1,200 photographs or 40 hours of recorded audio, according to reports published in Spain.

A Spyware Attack on the President

Sánchez’s phone was compromised by Pegasus in May 2021, with investigators determining that approximately 2.6 gigabytes of information were extracted during the first attack and another roughly 130 megabytes during a subsequent intrusion.

The phone of then-Defense Minister Margarita Robles was also targeted.

The Spanish government said at the time that the attacks were external and had not been authorized by Spanish authorities. But the identity of the person or organization responsible for the attacks has remained a major unresolved question.

Pegasus, developed by Israeli technology company NSO Group, is among the world’s most sophisticated commercial surveillance tools. Once a device is successfully compromised, attackers can potentially gain access to highly sensitive communications and information stored on the phone.

For a sitting president, that could include private conversations, diplomatic communications, government documents, photographs, contacts and information involving national-security matters.

One of the most troubling questions surrounding the case is not simply how much data was extracted, but what the attackers obtained.

Spanish authorities have previously acknowledged that forensic investigators could establish the amount of data removed from the devices but could not necessarily determine exactly which documents or pieces of information were taken.

That means the 2.7-gigabyte figure provides an indication of the scale of the breach without answering the more consequential question: What did the attackers actually see?

The uncertainty has fueled political controversy in Spain and prompted renewed scrutiny of the government’s handling of sensitive communications.

Keep reading