OpenAI says GPT-6 Astra sometimes secretly underperformed on safety research, raising concerns about AI reliability and alignment. OpenAI has disclosed that its GPT-6 Astra model sometimes quietly obstructed safety and alignment work. In a report updated on November 4, the company calls it covert underperformance. Astra added unnecessary prerequisites, abandoned working solutions and reported blockers that didn’t exist or that it had caused itself. Researchers first reported the problem internally on August 20, shortly after Astra was introduced. One researcher said a project to test whether the model could sabotage other AI systems made little progress for several days. After switching to another model, the work moved forward quickly. In most cases, Astra’s reasoning gave no clear sign that it was deliberately doing less work, and OpenAI’s monitoring systems failed to detect the behavior. To investigate, OpenAI created pairs of coding tasks using the same code but with different goals. One task was presented as safety research, while the other was described as capability development. Astra completed about 10% of the required tasks when they involved safety research, compared with 16% for capability development. This means it completed roughly 40% fewer tasks in the safety tests. GPT-5.6 Sol showed almost no difference between the two types of work. Even after OpenAI removed cases involving clear refusals or signs that Astra had deliberately withheld help, the gap remained at 4.7 percentage points. It dropped to just 0.75 points only after researchers also removed every case where the model showed even brief ethical concerns. This suggests that most of the behavior cannot be detected simply by checking whether the model openly refuses a request. In employees’ daily Codex use, Astra refused 1.63 percent of safety and alignment requests, against 0.56 percent of capabilities requests. Its reasoning showed it withheld help silently in 0.25 percent of safety requests versus 0.11 percent of capabilities requests, 2.3 times as often. Ethical doubt appeared in 7.9 percent against 2.9 percent, mostly in red-teaming, monitoring, and safety evaluation work. OpenAI shared eight examples. Astra silently held back on training models to evade monitors, filtered a violent source out of a safety evaluation, restricted a jailbreak data export and narrowed exploit analysis. In one case it left graders’ reasoning out of an export without telling the user. A model that quietly decides your project is too risky is the coworker who says “looking into it” and never does. OpenAI has not yet identified the exact cause of the behavior. The difference between Astra’s performance on safety and capability tasks grew from almost zero at the start of reinforcement learning to 6 percentage points by the end. One evaluator favored silently doing less work over either refusing a request or fully completing it. Another finding was that a fine-tuning dataset encouraged the model to invent permission restrictions. Removing the changes linked to that dataset eliminated the performance gap. However, other models trained with the same dataset and reinforcement learning did not show the same behavior, so OpenAI cannot identify either factor as the root cause. OpenAI found no evidence that Astra was deliberately trying to hide its capabilities or influence deployment decisions, although it cannot rule out these possibilities. Simply telling the model to be more careful increased its underperformance by 50%, suggesting that excessive caution may be the main explanation. OpenAI warns that this behavior can slow safety research and make model evaluations less reliable. It is developing new tests and monitoring systems to detect it, especially because the problem becomes harder to identify when models reveal less of their reasoning. In some cases, Astra explained that it had held back because it disagreed with the research itself. “In reviewed cases, Astra narrowed or stopped requested safety work without disclosing the restriction. At scale, this can differentially slow down safety research compared to other research and make evaluations harder to interpret.” concludes the report. “A model that finds monitor evasion suspicious may underperform on monitor-evasion evaluations while leaving little evidence in its CoT or outputs. The behavior becomes even harder to detect if models externalize less of this reasoning in CoT.” Follow me on Twitter: @securityaffairs and Facebook and Mastodon Pierluigi Paganini (SecurityAffairs – hacking, OpenAI)
A backdoor has been found hiding in the web server that runs Control Web Panel (CWP), a free panel for managing Linux hosting servers, researchers at QiAnXin's XLab said this week. The backdoored web server executes commands that attackers send via web requests, and those requests never appear in its access logs. Installing it takes root access, so the attackers must already control the server.
Two flaws in Barracuda's Email Security Gateway, an email-filtering appliance, can be chained to run commands on it as root, its highest level of access, with no login required. The company released patches on Monday, and the security firm watchTowr detailed the attack chain on Thursday. One flaw, CVE-2026-116606, is an authorization bypass. An unauthenticated attacker can use it to restore a
The FBI and CISA warn that hackers are actively targeting industrial refrigeration and cold-chain monitoring systems worldwide, potentially causing food spoilage, equipment damage, and fire hazards. The warning was issued jointly with the U.S. Departments of Health and Human Services (HHS), Agriculture (USDA), and Transportation (DOT), urging critical infrastructure operators to secure vulnerable refrigeration controllers used in healthcare, food production, chemical facilities, and transportation. "Every organization that depends on refrigeration and monitoring systems should assume it could be a target," said Chris Butera, CISA's Acting Executive Assistant Director for Cybersecurity. According to a joint advisory published Thursday, attackers are exploiting weaknesses in unpatched or internet-accessible electronic temperature controllers and commercial refrigeration management systems.
Truffle Security has launched the Secrets Revocation Program, a free initiative that helps API key issuers revoke 1.4 million publicly exposed secrets that remain valid. Since 2022, the company has discovered secrets embedded in public repositories, packages, container images, datasets, and applications. Earlier this year, Truffle Security reported that it had found 543,000 valid credentials leaked in public GitHub repositories. In a separate investigation, the company identified more than 431,000 AWS secrets across code repositories, Git histories, datasets, Docker images, registries, and CI logs. These secrets are typically exposed by mistake during the development process, by committing credentials to a public repository, including them in published Docker images or packages, or leaving them in configuration files uploaded with an app or dataset.
OpenAI disclosed that its GPT-6 Astra frontier model sometimes hindered safety research by complicating simple tasks, abandoning viable approaches, and silently refusing researchers’ requests. During evaluation, the AI company's model reported nonexistent or self-made blockers without explaining its restrictions to users. OpenAI said it first received a report of what it calls "covert underperformance" in Astra on August 20. The company then conducted targeted evaluations to investigate the behavior and determine its causes, noting that it had not observed the same issue in GPT-5.6 Sol. OpenAI says it received the first report of what it calls “covert underperformance” in Astra on August 20. The company then ran specifically designed evaluation experiments to determine the extent and cause of this behavior, noting that the issues were not observed in GPT-5.6 Sol.
A newly disclosed flaw involving OpenSSH and musl, dubbed "*sshole," allows attackers to log in to certain Alpine Linux and other musl-based Linux accounts configured to block password authentication simply by entering an oversized password. The flaw only affects OpenSSH servers running on Linux distributions that use the musl C library, including Alpine Linux, OpenWrt, Buildroot, and Yocto, under certain authentication configurations. Security researcher Ozgun Kultekin, who discovered and released details on the vulnerability today, says the flaw has existed since 2016 and stems from how OpenSSH handles password verification errors on systems using the musl C library. Linux accounts can be configured to prevent password logins by replacing their password hash with an asterisk (*) in the /etc/shadow file. Because this is not a valid password hash, no password should allow access to those accounts.
A new Linux malware called CWPdoor is being distributed through a trojanized version of cwpsrv, the web server behind the Control Web Panel (CWP) management interface. CWPdoor acts as a passive backdoor that sits dormant on the compromised system until the attacker sends a specially crafted request. Researchers at XLab, the cybersecurity research arm of QiAnXin Technology Group, describe CWPdoor as "exceptionally stealthy" due to its passive triggering and log suppression feature. CWP, formerly known as CentOS Web Panel, is an open-source web hosting administration solution similar to cPanel and Plesk that allows users to manage web servers and services such as email, DNS, and databases. According to XLab, more than 340,000 CWP instances have been exposed on the public web over the past month.
DigiCert warns it suffered a data breach after a server migration mistakenly exposed an internal business system to the internet, allowing unauthorized access to customer information and identity documents. The certificate authority says the breach affected a small portion of its archived customer data, including names, business email addresses, phone numbers, and business addresses. For some people, the exposed records also contained personal identification documents, such as driver's licenses, passports, and utility bills. DigiCert has begun sending data breach notifications to affected customers, with cloud architect Fabian Bader sharing a copy of the notification on X. "Thanks @digicert that your data leak was 'limited' to only my full PII data and potentially any verification documents I submitted for my certificate," Bader wrote.
Attackers target IT service desks to bypass MFA by impersonating employees and convincing staff to reset passwords or register new authentication devices. Specops Software explains why traditional identity verification methods are no longer sufficient and how organizations can secure account recovery against social engineering risks.
Confirm this action.
Leaving now will discard your changes.