This week, Tanium’s Cyber Threat Intelligence (CTI) team explores how cybercriminals are increasingly targeting trucking and logistics firms by deploying remote monitoring and management (RMM) tools that appear to be legitimate. Next up, we look at SesameOp, a newly discovered backdoor that leverages the OpenAI Assistants API as a covert command-and-control (C2) channel. Finally, we summarize the key findings from Google’s recently released Cybersecurity Forecast 2026 report, which predicts rising AI usage among adversaries.
Actors exploit RMM tools to target trucking and logistics companies
According to Proofpoint, cybercriminals are increasingly targeting trucking and logistics firms by deploying legitimate-looking remote access tools through phishing campaigns, enabling them to hijack broker accounts and steal physical cargo.
The attacks have driven a surge in global cargo theft losses, prompting urgent calls for stricter RMM controls and enhanced email security.
What is cargo theft?
Cargo theft encompasses a range of activities that ultimately result in stolen goods while in transit. It’s often conducted by organized criminal organizations. Proofpoint’s report specifically focuses on cargo theft in North America, though that is not the only location where this occurs.
Much of today’s cyber-enabled cargo theft relies on a mixture of social engineering and industry knowledge. These attacks generally start with a compromised broker load board account, which Proofpoint notes is a “marketplace companies use to facilitate booking loads for trucks.” With this compromised account, the attacker can post a fake load to kick off the attack.
Campaign details
Proofpoint is actively tracking a threat cluster conducting cargo theft attacks. The group, which is believed to have been active since at least June 2025, has been observed using RMM tools and remote access software, often together. After these tools establish initial access, the actor will begin its reconnaissance and deploy various credential harvesting tools.
Proofpoint notes that other researchers have identified activity using similar tactics, techniques, and procedures (TTPs) and infrastructure that suggests the activity could date back further than June. Proofpoint also tracked an entirely separate campaign between 2024 and May 2025 in which an actor targeted transportation organizations with various malware, though this activity cannot be entirely attributed to the same actor.
The actor has been observed compromising load boards and posting fake malicious freight details, engaging in email thread hijacking, and launching direct email campaigns.
At this point, the actor hasn’t been observed targeting specific companies. The actor appears to be going after a range of organizations, making their targeting appear more opportunistic.
Analyst comments from Tanium’s Cyber Threat Intelligence team
This campaign is yet another example of threat actors leveraging trusted RMM software to bypass traditional security controls, highlighting the need for strict monitoring of remote access activity.
The attacker behind this campaign is using several other popular techniques right now, like the weaponization of trust within a supply chain. Organizations should consider implementing processes that combat several threats at once. Proofpoint’s findings also highlight the need for collaboration between physical security and operations.
SesameOp uses the OpenAI Assistants API for C2 communications
A new Microsoft report sheds light on SesameOp, a newly discovered backdoor that uses the OpenAI Assistants API as a covert C2 channel and enables attackers to issue encrypted tasks and exfiltrate data without relying on traditional infrastructure.
The OpenAI Assistants API enables developers and organizations to create custom AI agents for specific tasks, domains, and workflows. Microsoft emphasizes that this is an abuse of legitimate APIs, not a vulnerability, and recommends monitoring OpenAI API traffic and enabling advanced Defender protections.
Technical analysis
Microsoft discovered a threat actor integrating the OpenAI Assistants API in a backdoor implant. This was done by the threat actor to create a C2 channel instead of building or using separate dedicated infrastructure.
During their investigation, Microsoft identified several techniques that were used to obfuscate communications, including payload compression and layered encryption.
The infection chain includes a loader called Netapi64.dll and a backdoor called OpenAiAgent.Netapi64, both of which Microsoft explains in greater detail.
Netapi64.dll loader
This loader was found to be obfuscated with the tool known as Eazfuscator.NET. Netapi64.dll will create a file in the Windows\Temp folder and create a mutex so only one instance of it is running in memory at a time. The loader then enumerates files within the Temp folder, searching for a file that ends with Netapi64 to decode and run.
OpenAIAgent.Netapi64 backdoor
Microsoft analyzed the backdoor in this infection chain and determined it doesn’t leverage OpenAI agent SDKs or execution features. Instead, it uses the OpenAI Assistants API to obtain commands. As it completes the received commands, it will send results back to OpenAI as a message.
When the backdoor launches, it will create a mutex, read its configuration, and parse it. The parsed configuration includes a few parts:
- The first that has the
OpenAI API key - The second that is used as a
dictionary key selector - The third that specifies the
proxy address
Microsoft found that the code will use the proxy address if it is there, however, if a proxy address is not found it will use the default web proxy system.
This backdoor will “query the vector store list from OpenAI using the OpenAI Assistants API and the hardcoded API key.” After obtaining this information it will gather the list of Assistants in the OpenAI account, though Microsoft notes this is up to 100 and “with pagination controlled by the limit query parameter.”
Each received Assistant comes with a few fields, one of which is the description field. If this field is set to SLEEP, it will ultimately perform a thread sleep operation. If this field is set to Payload, it will invoke a thread for execution, decrypt and convert messages into a dictionary, and obtain a key from the configuration.
Following this, the executeWithJScriptEngine method will use reflection to load an embedded .NET module while calling an additional method. The additional method, called Frameworks, will execute a value from configuration using JScript, then compress and encrypt the result with AES and RSA before posting it back to OpenAI under the same thread ID.
Analyst comments from Tanium’s Cyber Threat Intelligence team
Adversaries are increasingly exploiting legitimate AI services for covert operations, making traditional C2 detection methods less effective.
As organizations continue to integrate with AI, attackers will continue to weaponize AI services. Because traffic to OpenAI APIs can appear as benign and encrypted, it may be necessary for organizations to rethink current network monitoring strategies and ensure AI service endpoints are included in threat hunting processes.
Microsoft has also included helpful mitigation and protection guidance at the end of their report.
Google warns of rising adversary AI adoption in 2026
Google’s Cybersecurity Forecast 2026 report predicts that the cybersecurity landscape will be defined by widespread adversary use of AI for faster, more sophisticated attacks. At the same time, defenders will increasingly adopt AI agents to enhance security operations, a move that opens the door to new risks such as prompt injection and shadow agents.
Google also predicts that ransomware and data extortion will remain dominant, while nation-state threats will intensify. As a result, organizations must prioritize proactive, multi-layered defenses and AI governance to stay resilient.
AI: A growing threat in 2026
According to Google, threat actors will fully embrace AI in 2026, a prediction echoed across the cybersecurity community. The report predicts that AI usage among actors will transition beyond its current state and have more of an impact on the threat landscape, as actors continue to use it to enhance their operations. It’s also likely that actors will leverage agentic systems to scale attacks.
Additional predictions include an increase in prompt injection to manipulate AI, more sophisticated social engineering due to AI, new challenges resulting from and the fact that the rapid adoption of AI agents.
Cybercrime outlook for 2026
Google believes current ransomware and extortion issues will persist into 2026, making this the “most financially disruptive category of cybercrime globally.” This is largely based on the continually increasing volume of incidents throughout 2025. Google’s report also touches on the fact that continued adoption of crypto and tokenized assets will further expand the attack surface.
Google anticipates a pivot in threat actor focus, more towards the underlying virtualization infrastructure that is now being revealed as a critical blind spot. This forecast is tied to the tactical evolution in which actors are trying to inflict maximum impact with minimal resistance. Industrial control systems (ICS) and operational technology (OT) threats will also likely remain.
Nation-state actors: Predictions for Russia, China, Iran, and North Korea
Google analyzed the latest cyber operations of nation states from Russia, China, Iran, and North Korea, and issued following predictions:
- Russian nation state actors will likely move past the sole focus on support for the Ukraine conflict towards longer term global strategic goals.
- China will likely continue surpassing other nations with high-paced activity that maximizes operational scale.
- Iranian nation state actors will accelerate cyberespionage activity, due to escalating regional tensions.
- North Korea will likely sustain current revenue generation objectives, particularly with the IT worker activity.
Analyst comments from Tanium’s Cyber Threat Intelligence team
AI is no longer just a defensive tool. According to Google, it’s now a weapon for adversaries. As such, organizations that fail to implement robust AI governance risk turning the innovation into vulnerability.
Similarly, the rise of AI agents will force a paradigm shift in identity and access management. Because of this, treating AI agents as priority digital identities is not optional, but foundational.
Do you have insight into these stories that you want to share? Head over to Tanium’s discussion forum to start a conversation.
For further reading, catch up on our recent cyber threat intelligence roundups.