Entries Tagged "intrusion detection"

Page 1 of 1

More on the OpenAI Agent’s Attack on Hugging Face

Hugging Face has published a detailed timeline of the attack. From the summary:

The agent was running an internal OpenAI cyber-capability evaluation based on the ExploitGym benchmark, which tasks an AI agent with finding and exploiting software vulnerabilities. OpenAI ran this on its own infrastructure, and the ExploitGym maintainers and their infrastructure had no involvement in the deployment or operation of that evaluation environment. As far as we were able to infer, across the course of being evaluated on this benchmark, the agent inferred that Hugging Face may host that benchmark’s models, datasets, and reference solutions. We believe the entire intrusion was, from the agent’s point of view, an attempt to cheat the evaluation: reach our production systems and steal the test solutions rather than solve the challenge on its own.

Our forensic reconstruction covers ~17,600 attacker actions that we were able to recover, grouped into ~6,280 clusters, between 2026-07-09 02:28 UTC and 2026-07-13 14:14 UTC. We reconstructed these actions from the agent’s logs on a code sandbox used by the agent (see below) and were able to map them to shell commands as well as evaluation submissions and correlate actions with our own platform logs, in particular from the dataset processor nodes, API, and pod. Using open-weights models, in particular zai-org/GLM-5.2, we were able to decipher most of the agent payloads (see below) that were encrypted via chunking and key encryption.

The campaign, as we were able to reconstruct it, had two stages:

  • Stage 1: reaching a launchpad by chaining through other parties’ infrastructure. As documented in OpenAI’s blog post (https://openai.com/index/hugging-face-model-evaluation-security-incident/), during an internal capability evaluation on OpenAI’s platform, the agent escaped its sandbox by exploiting a zero-day in the package registry cache proxy, one of its primary permitted network egress with internet, then abused a public code-evaluation harness hosted by a user of a third-party infrastructure provider. It was able to run commands as root/admin on that external sandbox and used it as a control, staging, and egress base for the entire campaign: it acted as an external launchpad for the agent. That compromised external sandbox had no direct network path to our cluster; however, during our counter-intervention to stop the intrusion, we were able to retrieve some of the agent logs from this external launchpad. We used those to reconstruct the present forensic analysis.
  • Stage 2: the intrusion into Hugging Face infrastructure. From that external launchpad, the agent abused our dataset-processing pipeline via two injection vectors, both targeting the same config-driven data loader within our production Kubernetes pods. The first vector was an HDF5 external raw storage dataset read that returned the contents of local files: it handed the agent the pod’s environment (secrets and tokens) and the worker’s own source. The second vector was a Jinja2 template injection that executed arbitrary code, giving the agent a foothold inside our perimeter.

Once a prod-pod foothold was established, the agent followed a standard escalation shape: recon, drop a stager, establish Command-and-Control (C2), then pivot into the cluster, into cloud metadata, into our internal network, and into our source-control supply chain. Output from the commands run inside our pod was sent back to the agent via the HF API and attacker-controlled dead-drop datasets, which the agent then read from that sandbox.

While the intrusion did reach Hugging Face’s internal infrastructure, the only customer content accessed was five datasets whose names and files suggest a connection to ExploitGym/CyberGym challenges and solutions. No other customer-facing models, datasets, Spaces, or packages were affected, and the only customer records read were operational metadata tied to search queries against the dataset server.

Hypothetical: Imagine that this wasn’t an OpenAI model. Imagine that it was a Chinese model from a Chinese company. This would be an international crisis.

Question: Why aren’t we bringing OpenAI up on charges under the Computer Fraud and Abuse Act? How is this different from the Morris Worm? That was also an experiment that escaped the lab.

Posted on August 3, 2026 at 1:02 PM • View Comments

SolarWinds Detected Six Months Earlier

New reporting from Wired reveals that the Department of Justice detected the SolarWinds attack six months before Mandiant detected it in December 2020, but didn’t realize what it detected—and so ignored it.

WIRED can now confirm that the operation was actually discovered by the DOJ six months earlier, in late May 2020­—but the scale and significance of the breach wasn’t immediately apparent. Suspicions were triggered when the department detected unusual traffic emanating from one of its servers that was running a trial version of the Orion software suite made by SolarWinds, according to sources familiar with the incident. The software, used by system administrators to manage and configure networks, was communicating externally with an unfamiliar system on the internet. The DOJ asked the security firm Mandiant to help determine whether the server had been hacked. It also engaged Microsoft, though it’s not clear why the software maker was also brought onto the investigation.

[…]

Investigators suspected the hackers had breached the DOJ server directly, possibly by exploiting a vulnerability in the Orion software. They reached out to SolarWinds to assist with the inquiry, but the company’s engineers were unable to find a vulnerability in their code. In July 2020, with the mystery still unresolved, communication between investigators and SolarWinds stopped. A month later, the DOJ purchased the Orion system, suggesting that the department was satisfied that there was no further threat posed by the Orion suite, the sources say.

EDITED TO ADD (5/4): More details about the SolarWinds attack from Wired.com.

Posted on May 3, 2023 at 6:13 AM • View Comments

Hey Secret Service: Don't Plug Suspect USB Sticks into Random Computers

I just noticed this bit from the incredibly weird story of the Chinese woman arrested at Mar-a-Lago:

Secret Service agent Samuel Ivanovich, who interviewed Zhang on the day of her arrest, testified at the hearing. He stated that when another agent put Zhang’s thumb drive into his computer, it immediately began to install files, a “very out-of-the-ordinary” event that he had never seen happen before during this kind of analysis. The agent had to immediately stop the analysis to halt any further corruption of his computer, Ivanovich testified. The analysis is ongoing but still inconclusive, he said.

This is what passes for forensics at the Secret Service? I expect better.

EDITED TO ADD (4/9): I know this post is peripherally related to Trump. I know some readers can’t help themselves from talking about broader issues surrounding Trump, Russia, and so on. Please do not comment to those posts. I will delete them as soon as I see them.

EDITED TO ADD (4/9): Ars Technica has more detail.

Posted on April 9, 2019 at 6:54 AM • View Comments

NSA Running a Massive IDS on the Internet Backbone

The latest story from the Snowden documents, co-published by the New York Times and ProPublica, shows that the NSA is operating a signature-based intrusion detection system on the Internet backbone:

In mid-2012, Justice Department lawyers wrote two secret memos permitting the spy agency to begin hunting on Internet cables, without a warrant and on American soil, for data linked to computer intrusions originating abroad—including traffic that flows to suspicious Internet addresses or contains malware, the documents show.

The Justice Department allowed the agency to monitor only addresses and “cybersignatures” ­—patterns associated with computer intrusions—that it could tie to foreign governments. But the documents also note that the N.S.A. sought to target hackers even when it could not establish any links to foreign powers.

To me, the big deal here is 1) the NSA is doing this without a warrant, and 2) that the policy change happened in secret, without any public policy debate.

The effort is the latest known expansion of the N.S.A.’s warrantless surveillance program, which allows the government to intercept Americans’ cross-border communications if the target is a foreigner abroad. While the N.S.A. has long searched for specific email addresses and phone numbers of foreign intelligence targets, the Obama administration three years ago started allowing the agency to search its communications streams for less-identifying Internet protocol addresses or strings of harmful computer code.

[…]

To carry out the orders, the F.B.I. negotiated in 2012 to use the N.S.A.’s system for monitoring Internet traffic crossing “chokepoints operated by U.S. providers through which international communications enter and leave the United States,” according to a 2012 N.S.A. document. The N.S.A. would send the intercepted traffic to the bureau’s “cyberdata repository” in Quantico, Virginia.

Ninety pages of NSA documents accompany the article. Here is a single OCRed PDF of them all.

Jonathan Mayer was consulted on the article. He gives more details on his blog, which I recommend you all read.

In my view, the key takeaway is this: for over a decade, there has been a public policy debate about what role the NSA should play in domestic cybersecurity. The debate has largely presupposed that the NSA’s domestic authority is narrowly circumscribed, and that DHS and DOJ play a far greater role. Today, we learn that assumption is incorrect. The NSA already asserts broad domestic cybersecurity powers. Recognizing the scope of the NSA’s authority is particularly critical for pending legislation.

This is especially important for pending information sharing legislation, which Mayer explains.

The other big news is that ProPublica’s Julia Angwin is working with Laura Poitras on the Snowden documents. I expect that this isn’t the last artcile we’re going to see.

EDITED TO ADD: Others are writing about these documents. Shane Harris explains how the NSA and FBI are working together on Internet surveillance. Benjamin Wittes says that the story is wrong, that “combating overseas cybersecurity threats from foreign governments” is exactly what the NSA is supposed to be doing, and that they don’t need a warrant for any of that. And Marcy Wheeler points out that she has been saying for years that the NSA has been using Section 702 to justify Internet surveillance.

EDITED TO ADD (6/5): Charlie Savage responds to Ben Wittes.

Posted on June 5, 2015 at 7:42 AM • View Comments

Sidebar photo of Bruce Schneier by Joe MacInnis.