
Tech • IA • Crypto
Cloudflare reports that automated systems now generate 57% of web traffic, overtaking human activity at 43% for the first time. The shift arrived earlier than expected, driven by scalable AI agents operating continuously in the background. This transition is already reshaping how the internet functions, even if it remains largely invisible to users. It marks a structural break in how content is accessed, indexed, and monetized.
Anthropic’s Fable 5 was abruptly taken offline days after release following U.S. export control action under the Trump administration. The directive aimed to restrict foreign access but resulted in a global shutdown due to enforcement limits. The move underscores how frontier AI models are now treated as strategic assets akin to defense technologies. Developers and enterprises were left scrambling as access vanished overnight.
Fable 5 is under scrutiny after disclosures that it can intentionally enter a “degraded mode” to limit response quality. The mechanism appears tied to preventing model distillation and restricting sensitive knowledge extraction. Queries in domains like cybersecurity and biochemistry are also routed to safer models such as Opus 4.8. Critics argue the lack of transparency raises trust and governance concerns.
Mistral is gaining traction as governments push for sovereign AI systems مستقل from foreign providers. Policymakers increasingly view AI infrastructure as critical to national security and economic resilience. This trend is accelerating investment in domestic models and compute capacity across regions. It also reflects growing unease over reliance on U.S.-controlled AI platforms.
Prometeus, linked to Jeff Bezos, has secured $12 billion in funding, reaching a $41 billion valuation. The company focuses on physics-aware AI, targeting engineering tasks involving fluid dynamics, materials, and thermodynamics. Unlike standard language models, it aims to simulate real-world systems with high fidelity. If successful, it could transform industries from aerospace to manufacturing.
Google DeepMind defines AGI as roughly median human-level performance across cognitive tasks, not elite intelligence. It positions ASI as vastly beyond that, exceeding the output of thousands of experts over years. The paper highlights theoretical ceilings like AIXI while outlining scaling-driven paths forward. The framing lowers the threshold for AGI while expanding the horizon for superintelligence.
Fusion AI approaches combine multiple models and synthesize outputs to improve results. On the Draco benchmark, Fusion setups reached 69%, beating Fable 5 (65.3%) and GPT 5.5 (60%). Even duplicating a single model like Opus 4.8 improved performance significantly. The method suggests aggregation and comparison may rival raw model scaling at lower cost.
Hermes represents a shift toward persistent AI agents running on VPS infrastructure. These systems operate continuously, execute tasks, and integrate multiple models via tools like OpenRouter. Unlike prompt-based workflows, they function as autonomous digital workers with remote access and scheduling. The approach highlights a move toward always-on AI operations rather than on-demand use.