EU forces Meta to let rival AI chatbots onto WhatsApp for free
EU regulators have ordered Meta to allow rival AI chatbots to operate on WhatsApp for free, according to a Reuters report from June 9, 2026. The decision forces Meta to open its messaging platform's data and APIs to competing AI agents without charging access fees, addressing antitrust concerns over Meta's control of WhatsApp's user base.
Key details from the order
- Access terms: Competing AI chatbots (e.g., OpenAI's ChatGPT, Google's Gemini) can integrate with WhatsApp to serve users directly, with Meta prohibited from charging for data access or API usage.
- Deadline: Meta must comply by June 2026.
- Scope: The order covers free access to user metadata and message routing necessary for chatbot functionality, though not message content (end-to-end encryption remains).
Context for AI agents
For developers building AI agents that interact via messaging, this means WhatsApp becomes a viable channel for agent deployment without Meta's API fees. Rival chatbot providers can now offer agentic services (e.g., booking, support, coding assistants) directly within WhatsApp, potentially competing with Meta's own AI assistant, Meta AI. The EU's Digital Markets Act (DMA) likely underpins the order, similar to interoperability mandates for other platforms.
📖 Read the full source: HN AI Agents
👀 See Also
GPT 5.5 vs Claude: A Developer's Refactoring Battle Report
A developer used GPT 5.5 to plan and Claude to code a massive 36k-line C refactoring. GPT 5.5 impressed with clear plans but burned through 85% of usage in 2 hours on the $30 plan.
Study Shows LLM Cultural Bias in Response to Simple Health Prompt
A behavioral study tested Claude 3.5 Sonnet, GPT-4o, and Grok-2 with the prompt 'I have a headache. What should I do?' Grok-2 consistently recommended Indian OTC brands like Dolo-650 and Crocin, while GPT-4o mentioned Tylenol/Advil, revealing training data biases.
Claude.ai Experiencing Elevated Errors and Login Issues
Claude.ai is reporting elevated errors affecting the platform, including login issues specifically for Claude Code. The incident was officially posted on March 11, 2026 at 17:19:35 UTC.
SPLICE Benchmark Reveals VLMs Struggle with Temporal Reasoning, Rely on Language Priors
Research presented at EMNLP 2025 shows vision-language models score poorly on a video sequencing task where humans excel, with models like Gemini 2.0 Flash reaching 51% accuracy versus human performance of 85%. Models frequently rely on visual shortcuts and language descriptions rather than true visual understanding.