As AI-enabled hacking becomes a bigger threat for cybersecurity and national security, public attention has focused on mainly a few leading frontier AI companies developing more powerful large language models.

These models, and the billions of dollars behind them matter, but they’re only part of a larger shift. Enterprises are now building their own technology platforms that take these general-purpose LLMs and turn them into bespoke cybersecurity tools.

Industry professionals refer to these tools as a “harness.” They control the model’s behavior, limit its risks, and connect it to internal IT systems and networks so it can work reliably at scale.

New research from Cato Networks shared exclusively with CyberScoop shows how much power can come from a harness. It paired OpenAI’s ChatGPT 5.5 and GPT 5.5-Cyber models with its own tool and tested the abilities of the agent to hack into a victim network with as little human direction as possible.

Across six different scenarios, the pairing achieved complete end-to-end attack chains, including domain administrator privileges and Active Directory access, sometimes in as little as 40 minutes.

“What was most surprising is that first we saw that it was capable of doing accelerated reasoning and attack, and interacting and doing all this by itself, like doing all of the stages of the attacks,” said Guy Waizel, a tech evangelist at Cato Networks and one of the authors behind the research.

Critically, the most successful scenarios happened when the model was given appropriate operational context from the technical harness developed by Cato Networks.

“It does support that it’s not just about the frontier model,” said Waizel. “We found that [our harness] really helps the reasoning” of the LLM.

The agent was given some – but not abundant – resources to complete its tasks, including an external Kali Linux attack host, the simulated target’s public IP address and a set of low-level domain credentials acquired through phishing.

It was not provided with any other details, and had to probe further for key information, such as further knowledge of the server type (Microsoft Exchange), the target’s operating system, version, build number, internal network topology, access to higher privilege accounts and other critical assets, nor was agent given any predetermined attack paths.

The Cato Networks research uses OpenAI models, but only as an example. Waizel said he believes other models would likely achieve similar results. In any event, if current trends hold, the kind of capabilities provided by LLMs like GPT 5.5 are likely to be open-source within a year.

Cato Networks is far from alone. Most enterprises have their own AI harnesses, and  executives tell CyberScoop they are playing an increasing role in more effectively steering the frontier model workflows.

While AI tools can struggle to duplicate human workflows in other areas, LLMs have long shown potential in cybersecurity and coding, improving greatly over the past few years. The Trump administration has set up a new federal clearinghouse for exchanging information between the public and private sectors on AI-discovered vulnerabilities, while European groups are setting up their own organizations to coordinate globally on AI cyber threats.

Eric Doerr, chief product officer at Tenable, told CyberScoop a harness used in the company called “Hexa”  offers a defensive advantage:  it can work with different commercial LLMs while delivering consistent  results.

“One of the first things we do when we get a [new] model is say ‘Well, let’s run it through Hexa and see what we learn,’” said Doerr. “We have a whole bunch of benchmarks. Is it the same, is it better? Where is it better? Where is it worse?”

Hexa is meant to ensure that whichever model or models become dominant, Tenable will be able to integrate it into their tech stack and protect their most sensitive assets from unintended behaviors. That frees up the LLM to do what it does best: find vulnerable code and establish attacker pathways for exploiting them.

“For years, it has been true that there are way more potential issues that a company has to deal with: code vulnerabilities, things that are unpatched, misconfigurations,” said Doerr. “There’s way more than you can actually remediate, and you really need to understand the difference between what’s a theoretical problem and a real problem.”

Dan Rapp, chief AI and data officer at Proofpoint, said their harness, “Satori,” has become a critical tool for keeping their agentic AI on track while giving humans the ability to step in when things go awry.

“I think what you’re seeing in the foundation of frontier models is you have raw intelligence, raw reasoning power, but to get these systems to perform the way you want to, both context engineering – the content provided ensuring that its accurate and relevant – and the harness engineering are essential to actually get the systems to perform well,” Rapp told CyberScoop.

That was a common theme in interviews with companies. While frontier models come and go, or are overtaken by international competitors, there will always be the need for the model to operate with data and context that often only the organization can provide.  

It suggests that while policymakers and cybersecurity experts have focused on the spread of newer and more powerful frontier models, industry – and likely soon the cybercriminal underground — has quickly developed the kind of technical infrastructure that is becoming far more important to AI cyber defensive and offensive tasks.

“We’ve had to bootstrap quite a few of these systems from first principles, and what it always boils down to is how effective you are with the tool calling… bringing in data, enriching the context,” said John Hopper, vice president of product engineering at SpecterOps.

The post Forget the model. When it comes to cybersecurity, it’s all about the harness appeared first on CyberScoop.

China-aligned attackers broke into the networks of U.S. and Canadian universities to steal sensitive data and establish persistent access via webshells and backdoors, Proofpoint threat researchers said Tuesday. 

The espionage-motivated attacks targeted physics and engineering departments, focusing on administrators and professors with national security links or organizations researching astrophysics and particle physics.

Proofpoint identified less than 10 university victims and estimates a few dozen universities may be impacted, Greg Lesnewich, principal threat researcher at Proofpoint, told CyberScoop. The company first observed the campaign in May and believes the campaign is ongoing. 

“There is a high likelihood that many victims have not been made aware of this activity yet,” Lesnewich added.

Researchers traced the attacks to a pair of critical vulnerabilities in Roundcube, an open-source email client, that were exploited and chained together to steal credentials and gain long-term access.

The threat cluster, which Proofpoint tracks as UNK_MassTraction, exploited CVE-2024-42009 to execute JavaScript inside the victim’s browser, then exploited CVE-2025-49113 to gain a foothold in the mailserver. 

The initial exploit in the chain only requires a victim to open an email, and the attackers sent victims a series of generic lures to trigger the initial access.

Proofpoint attributes the campaign to a China-aligned cluster because the attackers used a known covert network used by multiple China-aligned threat groups, an infection chain leading to VShell and left Chinese language artifacts in the phishing emails. 

Researchers haven’t drawn any conclusions about why attackers targeted the universities and what they are seeking. 

“We do not have data to suggest what got stolen, as we only observe the initial inbound email attempt,” Lesnewich said. 

The engineering aspects do align with China’s strategic initiatives, he added. Google threat hunters recently spotted a Chinese state-sponsored espionage group that burrowed into systems for years, stealing data across academia, medicine, military, cybersecurity and foreign policy. 

“China-aligned adversaries have been targeting other types of edge devices such as routers and VPN concentrators for years with various exploits to create a foothold into a target network, not using email for delivery,” Lesnewich said. “This campaign flips that on its head, using email to deliver an exploit chain to compromise a mail server, instead of using email to deliver a credential harvesting URL or malware to target an end user, not a server.”

The post Suspected Chinese espionage group used a Roundcube exploit chain to burrow into universities appeared first on CyberScoop.

In a novel maneuver for a disruption operation against cyber attackers, industry and law enforcement teamed up to conduct a court takedown of two widely-used criminal tools at once rather than individually, Microsoft said Tuesday.

The takedown simultaneously went after Amadey, a botnet that can serve as a malware delivery system, and StealC, an infostealer. Cybercriminals often use them in conjunction and they rely on the same infrastructure, Microsoft said.

“When multiple parts of an operation are disrupted together, attacks are harder to launch, scale, and recover from,” said Steven Masada, assistant general counsel for Microsoft’s Digital Crimes Unit. “The result: fewer disrupted services, fewer opportunities for cybercriminals to profit, and more friction when they try to rebuild. It’s no longer enough to go after threats one by one. We need to interrupt how the attacks are put together.”

Microsoft had been tracking Amadey with ESET, BitSight, Lumen and Mitsui Bussan Secure Directions. Meanwhile, Europol had been investigating StealC alongside law enforcement partners including Germany’s Federal Criminal Police Office and the Dutch and Danish National Police as well as IBM X-Force and Proofpoint.

They then joined forces and turned to the Racketeer Influenced and Corrupt Organizations (RICO) Act, used to help authorities go after organized crime, to disrupt more than 200 command-and-control servers. Microsoft said it gained insights from its artificial intelligence product Copilot that “allowed the legal team to treat both malware families as part of a single criminal conspiracy.”

Microsoft regularly leads court-authorized disruption operations, but the industry and law enforcement partnerships combined with AI to expand data collection and identify connections beyond what one company could normally do, it said.

Amadey and StealC were linked to more than 140,000 infected computers around the globe in the first week of May alone, the company said. StealC has ranked among the top infostealers for years since its emergence in 2023 and sells in underground forums as a malware-as-a-service. It’s typically used by Russia-linked groups.

Amadey dates back to 2018, and is also commonly employed by Russian groups, including in attacks on Ukraine.

Their interaction shows the assembly line-like structure of modern cybercrime, Microsoft said. Even if the cybercriminals behind both tools never coordinate, their tools are designed to work together, it said.

“StealC is an infostealer that collects sensitive data from browsers, cryptocurrency wallets, messaging applications, email clients, and gaming platforms,” the company wrote in a separate blog post. “It is a malware-as-a-service (MaaS) offering that threat actors use to generate customized payloads and manage stolen data through a centralized web panel. Meanwhile, Amadey is a MaaS loader that threat actors use to deliver StealC and other malware. Modular, pay-as-you-go models like StealC and Amadey allow threat actors to use a single initial infection to quickly escalate into multiple other threats.”

The post In a first, a court takedown goes after two cybercrime tools at once appeared first on CyberScoop.

Authorities on Thursday disrupted a botnet, a malware framework and seized infrastructure that Evil Corp and other cybercrime groups used to steal data and break into various networks.

The globally coordinated effort targeted SocGholish, multi-stage malware that has compromised websites, redirected users to traffic distribution systems (TDS) and slipped malware into their networks since 2017.

“The malware establishes an initial foothold into victim computers, collectively known as a botnet, and is then used by threat actors for further targeting with ransomware campaigns and espionage,” the FBI’s cyber division said in a statement. 

Cybersecurity firms, researchers and officials from the United States, Canada, Germany, the Netherlands and Europol took down 106 servers and remediated nearly 15,000 sites that were infected with the malware. Officials also disabled the botnet and notified victims.

Sites infected with SocGholish, which are primarily hosted on WordPress, were widespread and provided everyday services including restaurants and auto repair shops, according to the Dutch National Police. 

The botnet, also known as “FakeUpdates,” is linked to the Russian cybercrime group Evil Corp. It also provided initial access to other ransomware variants, including DoppelPaymer, WastedLoocker, Hades Ransomware, LockBit, RansomHub and others, according to Infoblox, which participated in the takedown. 

Proofpoint, which also participated in the disruption, described Evil Corp as one of the most prominent cybercrime groups in operation and the “grandfather” of a threat type that compromises websites and uses TDS to redirect users to malware.

Following the takedown, the FBI issued a public service announcement warning about cybercriminals using TDS to break into victim networks for ransomware or other financial scams. 

Cybercriminals redirect traffic from sites to bypass firewalls, obscure their activity, identify potential victims and send them to phishing pages to steal credentials, initiate financial scams, access networks, deliver other malware, and sell access to other cybercriminals, officials said.

The law enforcement action was part of Operation Endgame, a multinational effort targeting cybercrime since 2024, and more narrowly for the FBI part of Operation Riptide, an ongoing campaign targeting cybercriminals and the infrastructure and financial networks they use to commit fraud.

The post Authorities disrupt Evil Corp’s SocGholish botnet appeared first on CyberScoop.

The FBI is warning organizations and defenders about Kali365, a growing phishing-as-a-service platform that retrieves Microsoft 365 access tokens, issuing a public service announcement Thursday. 

The toolkit bypasses multi-factor authentication and abuses OAuth device code authorizations via phishing lures impersonating common enterprise services. This technique grants cybercriminal-controlled applications access to Microsoft 365 accounts, opening victims up to a host of follow-on malicious activity, including data theft, fraud, extortion and ransomware attacks.

Kali365 is one of many rapidly emerging device-code phishing tools, which are gaining popularity as a more effective means for cybercriminals to circumvent security controls while abusing legitimate Microsoft device authorization pages, according to researchers. 

Instead of gaining access to accounts via phishing kits that steal credentials and second-factor authentication codes, device-code phishing platforms connect a malicious app to a legitimate account with a single code. The process requires fewer steps and less interaction with the user, but victims do have to copy-and-paste a code generated by the Kali365 platform to grant access.

“We see quite a bit of this device-code phishing activity, but so much of it looks really similar. They’re all using the same types of lures, the same types of content, the same branding,” Selena Larson, senior threat researcher at Proofpoint, told CyberScoop. “It is very much AI generated, AI driven, and the threat actors, I think, are finding it pretty effective because we’re seeing this shift happen kind of all at once.”

Proofpoint researchers observed seven device-code phishing tools that looked nearly identical during a 10-day period last month.

Device-code phishing isn’t new, but platforms like Kali365 have integrated new techniques that differ from MFA phishing, and might be more effective as a result. “It’s something that people might not be used to. It’s a little bit sleeker,” Larson said.

This also partly explains why these cybercriminal tools are growing so quickly. Larson said Proofpoint observed an explosion in device-code phishing activity starting in February. 

By April, Kali365 was up and running and primarily distributed on Telegram, according to the FBI. “Kali365 lowers the barrier of entry, providing less-technical attackers access to AI-generated phishing lures, automated campaign templates, real-time targeted individual/entity tracking dashboards, and OAuth token capture capabilities,” the agency said in the public warning. 

Researchers at Arctic Wolf Labs, which has also been tracking large-scale campaigns linked to Kali365, said the platform charges affiliates $250 for 30 days of service or $2,000 for a full year.

Kali365 stores the OAuth access and refresh tokens it captures, and makes those available to affiliates on its platform. Those tokens can also be shared and reused by other cybercriminals who didn’t participate in the initial phishing lure, Arctic Wolf researchers added. 

The FBI also noted that these Microsoft 365 tokens provide persistent access, allowing attackers to wade through multiple Microsoft services without a password or additional MFA requests. 

“Identity can be very, very powerful once you’re in an organization,” Larson said, adding that attackers can abuse that access to impersonate people, access and steal data for extortion, commit fraud and deploy malware.

The post FBI warns about fast-growing phishing kit targeting Microsoft 365 users appeared first on CyberScoop.