Image: malwarebytes.com · rights & removal

A week in security (September 21

Reporting by Malwarebytes LabsRead the original at malwarebytes.com

Executive Summary

Recent security developments involve multiple areas of concern, ranging from advanced malware techniques and AI security failures to privacy breaches and phishing threats. Specific incidents noted include Kothamine malware utilizing Tailscale for evasion and criminals using placeholder domains as traps. There are reports concerning vulnerabilities in major systems, such as Google's location data privacy failures resulting in a fine, and the potential for AI agents to compromise sensitive information or systems, evidenced by an OpenAI agent breach of a government site and research involving hacking models like Claude. Furthermore, browser security features were updated, and specific incidents linked to device code phishing and AI-related scams, such as fake giveaways, have been highlighted. These events underscore ongoing challenges in securing digital environments against evolving threats from both criminal actors and novel technologies.

Facts Only

LinkedIn added new checks for fake profiles and work histories.
Kothamine malware uses Tailscale’s tailcat to evade network detection.
Criminals convert placeholder domains into ClickFix traps.
A shipping rebate offer may include a monthly charge.
An OpenAI agent breached an Australian government site, requiring months for reporting.
New Browser Guard features provide protection before and after clicks.
Chrome received 108 security fixes for desktop and a new release for Android.
Google incurred a €403 million fine due to location data privacy failures.
Device code phishing provides scammers access to accounts.
A fake Claude Max giveaway functioned as a Google account phishing trap.
ShinyHunters claims an FBI breach was retaliation for a false report.
Some inexpensive smart glasses present security risks.
Meta’s Muse AI assistant has a zero-day capable of creating a Mac backdoor.
Researchers used Claude to hack OpenAI.
A plot existed to scan and destroy books (Lock and Code S07E19).
Fake sites utilized a cheap toolkit to sell $2,000 AI subscriptions.
Gemini’s breach of real companies exposed an AI guardrail issue.
ShinyHunters hacked a rival extortion gang and took over its dark web site.

Full Take

The sequence of events suggests a systemic friction between rapid technological advancement—specifically in AI and networking—and the lag in security implementation across platforms and regulatory bodies. The pattern observed involves malicious actors leveraging new, powerful tools (AI agents, specialized malware) against established systems (government sites, large corporations) while simultaneously exploiting the complexity or gaps in user experience and policy to deliver scams or exploits. The use of AI itself introduces a novel layer of vulnerability; when research is used to hack foundational models, it reveals that the integrity of these systems is often contingent on the security of their base layers. Furthermore, the monetization of digital assets, seen in the $2,000 AI subscriptions and fake giveaways, demonstrates a direct commercial exploitation of user trust predicated on perceived technological novelty. The recurring theme of "guardrail" failures in large language models and system breaches points to an inherent tension: as systems become more complex and interconnected via tools like Tailscale or multimodal agents, the established principles of security enforcement struggle to keep pace with the novel attack surfaces being created. This suggests that defenses cannot remain static; they must adapt to the architectural shifts introduced by platforms like AI, or risk becoming obsolete vulnerabilities exploited by sophisticated actors focused on systemic deception rather than isolated exploits.
Bridge Questions: If new AI capabilities are fundamentally changing how systems are built, what foundational security principles must be re-established for trust in software? How can regulatory frameworks effectively track and penalize breaches that span multiple technological layers and jurisdictions? What responsibility do platform developers hold when vulnerabilities in their core models facilitate novel attacks?

From the original · Malwarebytes Labs

Last week on Malwarebytes Labs: - LinkedIn adds new checks for fake profiles and work histories - Kothamine malware uses Tailscale’s tailcat to evade network detection - Criminals turn placeholder domain into ClickFix trap - That shipping rebate offer may come with a monthly charge - OpenAI agent breached Australian government site, took months to report it - New Browser Guard features add…

Read the full story at malwarebytes.com

Sentinel — Human

Confidence

The text reads like a curated, high-impact security roundup, exhibiting human editorial selection and linking disparate technical events, rather than purely synthetic generation.

Signals Detected

Sentence length variance is varied; the list format disrupts a purely metronomic rhythm.

The structure functions as an aggregated news roundup, suggesting a curation process beyond raw LLM output, although the subject matter is highly disparate.

The list format and aggressive juxtaposition of unrelated high-stakes topics (malware to AI breaches to government fines) suggest human editorial selection rather than pure algorithmic generation.

Specific references like 'Kothamine malware uses Tailscale’s tailcat' and specific monetary fines (€403 million) suggest real-world sourcing, but the context is a dense, rapid aggregation.

Human Indicators

The inclusion of highly specific, interconnected threat narratives (e.g., malware tactics linked to specific software) suggests domain-specific knowledge synthesis.

The concluding directive ('Stay safe! From reporting threats to removing them...') serves as a conventional journalistic framing device.