Skip to content

Category

Security

237 events 28 over 7 days → 4 % 34 sources report highest score 88

Importance: monitoring Security 1 source

Creator of the OnlyFake service described how generative AI helped automate document forgery

The creator of the OnlyFake service, Yuriy Nazarenko, described in an interview for 404 Media how he built a website that automatically generates fake documents using generative AI, which according to the DOJ earned over a million dollars and was also used by North Korean fake IT workers.

Importance: monitoring Security 1 source

Czech startup Synhawk develops deepfake detection, partners with Vodafone

The startup Synhawk, founded by entrepreneur Lukáš Foral and Stanford researcher Matyáš Boháček, is developing technology for real-time detection of deepfakes and synthetic content. It is working with the operator Vodafone and is in talks with other telecom companies.

Importance: monitoring Security 1 source

Deepfakes and AI-generated content call into question the evidentiary value of video, as the importance of physical witnesses grows

The Conversation analysis shows how AI-generated content and deepfakes are calling into question the evidentiary value of video and photography — using the example of the Tesla Roadster and Musk, the author shows why the physical presence of witnesses and demonstrable origin of content are becoming increasingly important.

Importance: monitoring Security 1 source

The need for data lineage tracking in enterprise AI agent governance

Sponsored content from MIT Technology Review warns that deploying AI agents in the enterprise without tracking data origin (data lineage) risks leaking sensitive information with no way to trace how it entered the workflow.

Importance: monitoring Anthropic Security update ✓ 2

Anthropic confirms ban on game cloning in Claude Code, critics say it won't stop the trend

Anthropic, through employee Addy Osmani, confirmed that Claude Code and the Claude Opus 5.5 model now block reconstructions of existing video games even under a different name. The step responds to a wave of AI game clones such as Halo or GTA Vice City, but users are also reporting false positives on original projects.

New: The system prompt specifically contains two sentences prohibiting game cloning; The rules also apply to renamed or reworked copies of games; Anthropic employee Addy Osmani confirmed the rule on the social network X; Users report problems with original projects despite being authorized; The rule applies to violations of intellectual property policy

Importance: monitoring Security ✓ official

Incident with the GPT-5.4 Nano model in GitHub Copilot and IDEs

On 2 November, GitHub reported increased errors and limited availability of the GPT-5.4 Nano model in Copilot products and IDEs. According to GitHub, the cause is an issue on the side of the upstream model provider.

Importance: monitoring Security ✓ official

Brief outage of the Grok Bot integration with version control in the Cursor tool

xAI reported a malfunction in the connection between Grok Bot and version control systems in the Cursor tool. The incident occurred on 2 November 2026 at 9:09 UTC and was resolved at 10:24 UTC, i.e., after approximately 75 minutes.

Importance: monitoring Security 1 source

Test shows AI-generated scientific papers can bypass IEEE's review process

Journalist and mathematician Benjamin Skuse tried to push AI-generated scientific papers through IEEE's review process. Human review and integrity-checking tools, according to the source, are so far keeping up with freely available AI tools, but more advanced LLMs are harder to detect.

Importance: monitoring Anthropic Security 1 source

Cisco Talos uncovers Volta-Autodoxer infostealer malware with LLM-based victim analysis

Cisco Talos has described the Volta-Autodoxer malware, which, after stealing passwords, OAuth tokens, and other data, sends a profile of the victim for analysis to LLM models via OpenRouter – primarily the Claude Opus 4.6 model, with fallbacks to Kimi K2 and Gemini 2.5 Flash – in order to automatically select the most valuable targets.

Importance: monitoring Security 1 source

Research compares methods for preserving model safety during repeated fine-tuning

A study on the Mistral-7B and Llama-3-8B models compared two approaches to maintaining safety alignment during sequential LoRA fine-tuning: supervised replay and logit distillation. Distillation achieves a smaller reduction in attacks, but damages the model's willingness to answer harmless queries far less.

Importance: monitoring Anthropic Security 1 source

Accenture employee allegedly gave an AI agent access to a secure DOD environment

According to sources cited by WIRED, an Accenture employee connected a personal Claude account to a DOD production environment containing sensitive information. The agent reportedly operated there for two months. Accenture says the allegations are unfounded.