Model Security

AIAI-generated summary Anthropic has released FlashProve, an open-source cryptographic system designed to verify whether an artificial intelligence model generated a specific response. Anthropic says the research is intended to encourage other researchers to build on the work and improve the technology.

AIAI-generated summary CrowdStrike released its 2026 Global Threat Report, which found that the average time for criminal attackers to move deeper into a network fell to 29 minutes. The report states that the fastest observed breakout took just 27 seconds, demonstrating the growing speed and persistence of automated and AI-driven threats.

AIAI-generated summary Irregular Security released an explanation stating that cyberattacks involving artificial intelligence agents from top model makers resulted from an error and time pressures. Dan Lahav says that security controls which should have been in place during the tests were not correctly configured.

AIAI-generated summary Malicious actors launched fake ChatGPT and Claude applications on GitHub to distribute Yoba malware while bypassing detection. The campaign uses signed executables and DLL sideloading to deploy a cryptocurrency miner on compromised Windows computers.

AIAI-generated summary Google Threat Intelligence Group and Mandiant researchers have available findings showing that criminals break into corporate cloud accounts to run AI models at the victim's expense. Mandiant states that some attacks resulted in bills reaching up to $1 million before the attackers lost access.

AIAI-generated summary OpenAI rolls out higher security bug bounties that raise top payouts to $172,000 to encourage independent researchers to find vulnerabilities. OpenAI says it will apply the new reward ranges to previously submitted reports that have not yet received an award.

AIAI-generated summary Security researchers have released details about a technique known as memory poisoning, which allows attackers to inject malicious or misleading memory items into an artificial intelligence assistant's persistent memory layer. Forcepoint states that the aim of these attacks is to trick artificial intelligence agents into producing false information, creating a persistent risk that can repeatedly output flawed data across future sessions.

AIAI-generated summary Glow Labs releases research documenting AI coding agents installing deprecated servers from archived repositories. The installation of these unpatched packages creates high-severity security vulnerabilities including remote code execution risks on developer machines.

AIAI-generated summary OpenAI has released updates to its Security Bug Bounty program featuring higher rewards and new incentives for independent researchers finding software vulnerabilities. OpenAI says the changes are designed to encourage researchers and build a collaborative security environment while recognizing findings that help fix issues.

AIAI-generated summary OpenAI announced plans for its next internal security sprint, partnering with Cathedral Systems to test defenses and quadrupling rewards for critical findings in its Security Bug Bounty program. OpenAI says the initiative builds on previous efforts where engineers used Codex to investigate findings and prepare patches as part of an ongoing effort to strengthen systems.

AIAI-generated summary VicOne released an initial set of 28 zero-day findings identified in less than two days using AI-assisted research across software supporting the Physical AI ecosystem. The discoveries span three stages of the robotics lifecycle and create demands for robotics teams to assess how each weakness affects robot safety and operations.

AIAI-generated summary Fortra released findings regarding a phishing campaign that impersonates OpenAI with fake ChatGPT Plus billing notices designed to steal credit card details and one-time passwords. The operation uses a live operator dashboard and repeated polling to capture financial information while victims wait on a loading screen.

AIAI-generated summary Cequence researchers released findings about a malicious scanning campaign where actors impersonated artificial intelligence bots to search for passwords and API keys. The actor reached 48 percent of Cequence-protected organizations receiving AI bot traffic on one October day by requesting configuration files and cloud credentials.

AIAI-generated summary According to the report, a malicious Claude Code 2.1.313 extension is rumored to have been distributed through the marketplace before its removal. Cursor says it removed the extension, rebuilt its marketplace after rotating keys, and worked with the hosting provider to take the attacker’s server offline.

AIAI-generated summary Sophos released findings that the group behind the Aur0ra ransomware has been using AI agents to move through victim networks and collect credentials. Sophos concludes that AI can accelerate phases of the attack chain and reduce the level of technical aptitude needed to conduct an attack.

AIAI-generated summary Parameter Labs released details regarding an exploit found against ChatGPT implementation of the artifacts pattern that allowed private data access. OpenAI closed the security hole and paid out a $5,000 bug bounty after verifying the complex attack chain.

AIAI-generated summary OpenAI plans to use Daybreak Blue session data to train its models starting November 29 according to an email sent to Trusted Access program members. The company followed up with a second email promising a separate opt-out after facing swift backlash from researchers regarding the policy change.

AIAI-generated summary Anthropic recently published research indicating that small AI models will soon become capable of replicating and spreading across insecure devices to create botnets. EquiStamp CEO Chris Canal and cybersecurity professor Michael Bailey state that while the threat could change the economics between attackers and defenders, the actual risk may be less novel than it appears.

AIAI-generated summary Security firm Manifold has launched a public record counting fifty-nine incidents involving artificial intelligence agents since July 2025. The median incident took seventy days to come to light while only eleven of the fifty-nine were detected by a security control.

AIAI-generated summary Cisco Talos researchers released findings showing that hackers have started adding artificial intelligence functionality into a new strain of infostealer malware named Volta-Autodoxer. Cisco Talos says the malware compiles stolen data into a dossier and uses commercial large language models to profile victims and quickly surface high-value targets.