Happy week 11!
Before we jump in you might have missed my post:
The Best LLM for Cyber Threat Intelligence: OpenAI vs Anthropic vs Groq
We compare AI services and LLMs for cyber threat intelligence to find the best one for speed, context and cost.
Nikoloz Kokhreidze
This week in cybersecurity, critical vulnerabilities in ChatGPT's plugin ecosystem were discovered, Roku disclosed a data breach impacting over 15,000 customers, and the Tor Project introduced WebTunnel to bypass censorship.
Let's dive in!
🌐 This Week in Cybersecurity
Roku Discloses Data Breach Impacting Over 15,000 Customers
- Credential Stuffing Attack: Roku disclosed that 15,363 customer accounts were compromised in a credential stuffing attack, where threat actors used credentials exposed in previous data breaches to gain unauthorized access to Roku accounts.
- Account Takeover and Fraudulent Purchases: Once an account was breached, the attackers changed account information like passwords, email addresses, and shipping addresses, effectively locking out legitimate users. They then used stored credit card details to make fraudulent purchases of hardware and streaming subscriptions without the account holders receiving order confirmations.
- Stolen Accounts Sold on Dark Web: Threat actors have been conducting these credential stuffing attacks for months using tools like Open Bullet 2 and SilverBullet. Successfully hacked Roku accounts are being sold on stolen account marketplaces for as little as $0.50 each. Buyers then hijack the accounts and make illegal purchases using the stored payment methods.
- Roku's Response and Lack of 2FA: Upon detecting the breach, Roku secured the impacted accounts, forced password resets, and refunded unauthorized charges. However, the platform does not currently support two-factor authentication (2FA), which could prevent account takeovers even if credentials are compromised. Roku also recently updated its Dispute Resolution Terms, possibly in part due to the ongoing credential stuffing attacks and fraud.
ChatGPT Plugin Vulnerabilities Allow Access to User Accounts and Sensitive Data
- Overview: Salt Labs researchers discovered critical vulnerabilities in ChatGPT's plugin ecosystem that could allow attackers to access user accounts and sensitive data on third-party websites like GitHub without user interaction. The flaws were found in ChatGPT itself and in multiple plugins developed using the PluginLab.AI framework.
- ChatGPT Vulnerability: A flaw in ChatGPT's plugin installation process allowed attackers to install malicious plugins on victim accounts without their approval. By exploiting this, attackers could forward almost any chat data, including credentials and sensitive information, to their own plugin. ChatGPT implemented a state parameter but it was not sufficiently random.
- PluginLab.AI Framework Flaws: Dozens of plugins developed with PluginLab.AI, such as AskTheCode, were vulnerable to a zero-click account takeover. Unauthenticated requests could be made to retrieve a victim's member ID and generate a code to access their GitHub repositories via the plugin. PluginLab.AI quickly mitigated the issue after disclosure.
- OAuth Redirection Manipulation: Several plugins, like Charts by Kesem AI, had a classic OAuth vulnerability allowing attackers to provide a malicious redirect URI and steal user credentials. Proper validation of the redirect URI was lacking. Documentation improvements emphasizing redirect URI security were recommended to OpenAI.