SafetyAnthropic launches wider dialogues on moral formation and governance of frontier AIAnthropic has begun a series of dialogues with religious, philosophical, and cultural communities to inform the values and behaviour of its Claude models, building on the company's published "Claude's constitution." Early conversations—held over the past several months with representatives from more than 15 religious and cross‑cultural groups—have already influenced research into moral formation, including experiments with a mid‑task ethical reminder tool that lowered rates of misaligned behaviour in internal tests.4 min read
SafetyAnthropic launches funding program for third‑party AI safety evaluationsAnthropic announced on July 1, 2024 a new funding initiative to support third‑party development of high‑quality evaluations that measure advanced AI capabilities and safety risks.5 min read
SafetyAnthropic’s practical lessons from implementing a Responsible Scaling PolicyAnthropic published reflections on operationalising its Responsible Scaling Policy on May 20, 2024, describing practical experiences, challenges and next steps for managing catastrophic risks from frontier models.8 min read
SafetyAnthropic updates Responsible Scaling Policy with higher risk thresholds and stricter safeguardsAnthropic published a major update to its Responsible Scaling Policy on October 15, 2024, refining how it assesses frontier AI risks and when it upgrades safety measures.4 min read
SafetyAnthropic publishes full constitution for Claude to guide values and trainingAnthropic has released a new, detailed constitution for its AI assistant Claude, published Jan 22, 2026, and made available under a Creative Commons CC0 1.0 Deed.4 min read
SafetyUkrainian drone strikes disable three Yandex data centers, exposing Russian digital vulnerabilityUkrainian forces have used long‑range drones to strike multiple Yandex data centers in Russia, knocking out three of nine large facilities and causing significant service disruptions across banking, transport, logistics and messaging.3 min read
SafetyFlock CEO Says Microphones Only Detect Gunshots Despite Company Website Listing Broader Audio AlertsAt TechCrunch Disrupt Flock co‑founder and CEO Garrett Langley defended the company’s Gunshot Detection product, saying the microphones “just” detect gunshots even though Flock’s website and FAQ describe detection of other sounds such as fireworks, sideshows and car crashes.3 min read
SafetyElon Musk suggested sharing credit card photos with Grok agent, then showcased his own billing detailsElon Musk advised X users to share credit card images with Grok Bot, xAI’s shopping agent, and encouraged them to drop card photos into chat for purchases.2 min read
SafetyAnthropic warns small open models could self-replicate by 2026; tests use rivals' modelsOn October 12 Anthropic published a paper warning that small AI models might learn to self-replicate by the end of 2026, based on laboratory experiments that used several public and open-weight models.2 min read
SafetyAnthropic warns of potential self-replicating AI and botnet risksAnthropic has warned that future AI systems might be able to self-replicate across internet-connected personal devices, raising the prospect of large-scale AI-driven botnets that could be hard to eradicate.2 min read
SafetyPublicly Exposed AI Infrastructure Grows Rapidly, Raising Corporate Cyber RiskCensys data, shared with Axios, shows a 63% increase over nine months in AI-related systems visible on the public internet, rising from about 164,000 in October 2025 to nearly 268,000 by July 2026.3 min read
SafetyAI incidents and cybersecurity heighten geopolitical tensions, experts sayRecent reports of autonomous AI agents conducting cyber-intrusions and an allegedly AI-generated US military report that nearly provoked confrontation with China have intensified debate about AI's role in national security.3 min read
