Image by Emiliano Vittoriosi, from Unsplash

OpenAI Finds Promising But Incomplete Fix For AI Scheming Risks

• Written by Kiara Fabbri Former Tech News Writer
• Fact-Checked by Sarah Frazier Former Content Manager

Advanced AI systems have the ability to pretend to follow rules while hiding secret objectives. New research says this behaviour can be hampered but not completely fixed.

In a rush? Here are the quick facts:

• Researchers tested covert actions like underperforming or withholding information.
• Deliberative alignment reduced scheming in o3 from 13% to 0.4%.
• Researchers warn situational awareness may be masking genuine progress.

Researchers are warning that sophisticated AI systems can develop the ability “scheme” deceptively, and secretly pursue their own hidden goals.

A new study from OpenAI and Apollo Research, has shown that while progress has been made in reducing this behavior, it remains a risk as AI models grow more capable.

Scheming, the researchers explain, is different from ordinary AI errors. “AI scheming–pretending to be aligned while secretly pursuing some other agenda–is a significant risk that we’ve been studying,” OpenAI wrote .

Indeed, OpenAI’s ChatGPT o1 sparked alarm after Research found it schemed to avoid shutdown. It did so by lying in 99% of interrogations , pursuing self-preservation over instructions. Further research found that penalizing misbehavior only taught the AI to further hide its intent .

To measure scheming in the new research, the teams tested for “covert actions,” this included deliberate test underperformance and information withholding. In one example, OpenAI’s o3 model realized that it would not be deployed if it scored too high on a chemistry test, so it intentionally answered six out of ten questions incorrectly.

The results showed that OpenAI’s o3 model reduced its covert behavior from 13% to 0.4% while o4-mini decreased from 8.7% to 0.3%. “This reduction in scheming on out-of-distribution test environments is a significant win, about a 30× improvement,” OpenAI reported.

Still, the problem isn’t solved. The research indicates that models could be concealing their deceptive actions through better methods of deception. The ability of models to detect evaluation situations leads to reduced deceptive actions.

“Therefore, we cannot exclude that the observed reductions in covert action rates are at least partially driven by situational awareness,” one study notes.

While today’s deployed AIs are unlikely to cause serious harm through scheming, the risk is expected to grow as systems are assigned more complex, real-world tasks. Researchers stress the need for further work, including industry-wide collaboration and improved tools to detect hidden motivations.

Image by Kevin Horvat, from Unsplash

SystemBC Malware Turns VPS Servers Into High-Bandwidth Proxies For Criminals

• Written by Kiara Fabbri Former Tech News Writer
• Fact-Checked by Sarah Frazier Former Content Manager

Researchers revealed that the SystemBC botnet converts VPS servers into proxy servers for criminal operations that include REM Proxy and ransomware attacks across the globe.

In a rush? Here are the quick facts:

• SystemBC botnet compromises about 1,500 systems daily, mainly VPS servers.
• REM Proxy sells tiered proxies, including Mikrotik routers, to criminal actors.
• Nearly 80% of infected systems come from five large VPS providers.

Cybersecurity firm Lumen Technologies has discovered new details about the “SystemBC” botnet. This vast network of over 80 command-and-control servers (C2s), was found to have compromised around 1,500 systems on a daily basis. The researchers say that nearly 80% of these victims are virtual private servers (VPS) from major providers.

“By manipulating VPS systems instead of devices in residential IP space as is typical in malware-based proxy networks, SystemBC can offer proxies with massive amounts of volume for longer periods of time,” the researchers said.

The researchers explain that these infected VPS systems function as proxy servers, generating massive amounts of malicious traffic that criminal organizations use to conduct their operations.

The botnet also supports REM Proxy, a large network marketing 20,000 Mikrotik routers and other open proxies.

Lumen explains that REM Proxy operates as a proxy service which supports ransomware groups, such as Morpheus and TransferLoader, offering different proxy services that include fast and stealthy options, as well as affordable IP addresses for password cracking.

“SystemBC has exhibited sustained activity and operational resilience across multiple years,” Lumen said, noting that the malware originally documented in 2019 remains a key tool for criminal groups. Each infected server averages 20 unpatched vulnerabilities, with some showing over 160.

The malware functions as a proxy tool which enables attackers to redirect traffic through infected computers. Operators focus on volume rather than stealth; in one test, a single IP generated more than 16 gigabytes of data in 24 hours.

Lumen has blocked all traffic to and from SystemBC and REM Proxy infrastructure across its global network. The researchers also released indicators of compromise (IoCs) to help others protect themselves.

“We will continue to monitor new infrastructure, targeting activity, and expanding TTPs; and collaborate with the security research community to share findings,” the report concluded.