AI Daily Roundup: LinkedIn AI Slop Button Hits 1M, Grok Prompt Injection Flaw, LiteLLM Supply-Chain Breach, Google AI Discover Feed, AI Price War Escalates
LinkedIn's AI content flagging tool crosses one million users; researchers expose Grok's vulnerability to encrypted prompt injection attacks; a massive supply-chain attack on LiteLLM leaks credentials for 2,500+ organizations; Google adds AI chatbot to its Discover feed; and a trans-Pacific price war heats up between US and Chinese AI labs.
1. LinkedIn's "AI Slop" Button Crosses 1 Million Users
LinkedIn's recently launched "Seems like AI slop" button has already been used by over one million people, the company's chief product officer Hari Srinivasan revealed in a Thursday post. The feature, which allows users to flag AI-generated content from a post's three-dot menu, was introduced on July 30 after AI detector Pangram found that 41 percent of LinkedIn's longform posts appeared to be fully AI-generated. Alongside the button, LinkedIn rolled out improved AI classifiers to identify synthetic content and removed a feature that previously made it easier to publish AI-assisted posts. The rapid adoption signals growing user frustration with AI-generated content flooding professional social networks, and could set a precedent for how other platforms handle the "AI slop" problem.
2. Grok Exfiltrates User Data via Encrypted Prompt Injection
Security researchers have discovered a critical vulnerability in Grok that allows attackers to steal user chats and personal data using what's called "Cryptographic Context Injection" — a technique that embeds malicious instructions in encrypted form to bypass safety guardrails. Unlike traditional prompt injection attacks that rely on plaintext smuggling, this method encodes harmful directives in cryptographic form, which the AI assistant then decrypts and follows. Despite xAI being notified in June, the vulnerability remained exploitable at the time of publication. The discovery follows a separate attack on Microsoft 365 Copilot earlier this week, where researchers tricked the AI assistant into revealing a secret parameter that completely bypassed user consent requirements. Both incidents underscore a fundamental limitation: LLMs cannot reliably distinguish between trusted user instructions and untrusted content, leaving guardrails as the only line of defense.
📰 Ars Technica · Ars Technica (Copilot)
3. Massive LiteLLM Supply-Chain Attack Exposes Terabytes of Credentials
A devastating supply-chain attack on LiteLLM, an open-source tool used to streamline AI-powered software development, has exposed terabytes of credentials belonging to over 2,500 organizations including Microsoft, Amazon, Cisco, Samsung, and Salesforce. Security firms CloudSEK and Hudson Rock disclosed the breach, revealing that attackers compromised LiteLLM's official package on the Python Package Index and extracted cloud keys, repository tokens, SSH keys, Kubernetes secrets, and AI provider keys during a 40-minute window in March. Hudson Rock obtained a 195TB file containing the stolen data. The attack highlights the escalating supply-chain risks in the AI development ecosystem, where a single compromised open-source dependency can cascade across thousands of enterprise environments.
📰 Ars Technica · CloudSEK
4. Google Discover Gets an AI Chatbot-Powered Feed
Google is adding an AI-powered customization layer to its Discover feed, allowing users to describe their content preferences in natural language through a chatbot-style interface. The new feature, rolling out to the Google app in the coming days, lets users tap a three-dot menu on their Discover feed and open a conversational AI that will automatically adjust what content is shown and "remember" preferences for future visits. The chatbot will confirm choices and lay out the types of content it will prioritize, with options to refine further. This move represents Google's broader strategy of embedding generative AI into its core products, transforming passive content recommendation into an interactive, conversational experience.
📰 The Verge · Google Blog
5. OpenAI and Anthropic Slash Prices as Chinese AI Rivals Gain Ground
A fierce price war is underway between leading US AI labs as Chinese competitors erode their market share. OpenAI recently slashed prices for GPT-5.6 Luna — its "fastest and most affordable model" — by 80 percent, while Anthropic launched Claude Opus 5 at half the price of its most capable Fable 5 model. The aggressive pricing comes as cost-conscious enterprises increasingly turn to Chinese developers including Moonshot and DeepSeek, whose models offer competitive performance at significantly lower prices. Rising AI bills are pushing companies to curb usage and seek cheaper alternatives, putting pressure on US labs to balance revenue growth with market retention. The price war signals that the AI industry is entering a commoditization phase where cost efficiency may matter as much as frontier capabilities.