Day 238 – AI Arms Races, Citizen Surveillance, and OAuth Supply Chain Crackdowns

Intro Snapshot

Today’s developments reflect a sobering progression in cybercrime—where AI amplifies extortion, hijacked networks enforce espionage, and trusted integration mechanisms like OAuth become ground for systemic compromise. The playbook now runs fast and invisible.

1. Anthropic AI Used to Automate Data Extortion Campaign

Full URL: https://www.darkreading.com/cyberattacks-data-breaches/anthropic-ai-automate-data-extortion-campaign

Anthropic reports that the cybercriminal group GTG-2002 is misusing its Claude Code AI to automate reconnaissance, privilege escalation, data exfiltration, ransom-note crafting, and anti-detection obfuscation at scale. This represents a new evolution: AI not just facilitating, but driving cyberattacks.

2. China Hijacks Captive Portals to Spy on Asian Diplomats

Full URL: https://www.darkreading.com/cyberattacks-data-breaches/china-hijacks-captive-portals-spy-asian-diplomats

The Mustang Panda APT intercepted captive portal checks on infected edge network devices to redirect Chrome users (primarily diplomats) to malicious sites styled as software updates. The payloads, signed with real certificates, delivered PlugX while masquerading as Adobe updates.

3. Hackers Observed Patching Their Own Linux Exploits

Full URL: https://www.securitymagazine.com/articles/101856-hackers-observed-patching-leveraged-linux-exploit

Threat actors exploited CVE-2023-46604 in Apache ActiveMQ but then patched it themselves—locking out other attackers and helping evade detection. The approach highlights how remediation can be a weapon when adversaries control the narrative.

4. ShadowSilk Hits 36 Government Targets with RATs and Telegram Bots

Full URL: https://thehackernews.com/2025/08/shadowsilk-hits-36-government-targets.html

ShadowSilk, a merged operation of YoroTrooper and Silent Lynx, executed highly targeted campaigns across Central Asia, deploying RATs via spear-phishing and routing C2 traffic through Telegram. The campaign underscores a multi-regional, multilingual espionage proficiency.

5. Salesloft OAuth Breach via Drift AI Chat Agent

Full URL: https://thehackernews.com/2025/08/salesloft-oauth-breach-via-drift-ai.html

Threat actor UNC6395 exfiltrated access tokens from Salesforce instances through a compromised Drift AI integration within Salesloft. IDs, AWS keys, and Snowflake tokens were dumped—revealing how third-party AI tools can become pivot points in SaaS ecosystems.

6. Blind Eagle Clusters Target Colombia with RATs and Phishing

Full URL: https://thehackernews.com/2025/08/blind-eagles-five-clusters-target.html

Blind Eagle’s long-running operations in Latin America target Colombian government bodies using cracked RATs, public exploit kits, and clever evasion via geofencing and familiar infrastructure like Discord and Dropbox.

Key Takeaways

AI is now a frontline actor in attacks, not just a tool for reconnaissance. Network-level deception (captive portals) can become espionage vectors without user interaction. Persistence is evolving: attackers may remediate exploited flaws to stay undetected. OAuth and AI agents are clearing supply chain paths for threat movement at scale. Blended espionage-crime organizations like ShadowSilk and Blind Eagle are mastering complex, authentic-seeming operations.