Modelos, agentes, investigación y negocio de la IA.
Instacart has drastically reduced its technical debt by delegating code generation to AI agents. Its CTO, Anirban Kundu, revealed that in 97% of cases developers no longer read the generated code, and idle systems are simply discarded and rebuilt. This marks a paradigm shift: technical debt ceases to be a problem when code is constantly regenerated.
Microsoft and Wiz have unveiled bug-hunting systems based on multiple AI agents that detect over 90% of vulnerabilities, surpassing individual models like GPT-5.5 or Claude Opus. The key is using the right model for each task, cutting costs in half.
OpenAI has quietly modified ChatGPT to refuse to imitate the style of famous authors. The measure, detected by Ars Technica, responds to legal pressure from creators who claim unauthorized use of their works. We explain what changed, why it matters, and what it means for the future of generative AI.
Hundreds of Claude conversations were indexed on Google due to missing noindex tags. Personal data and credentials were exposed. Anthropic fixed the error, but the case highlights persistent privacy risks in AI.
Snowflake has launched Cortex AI Gateway, a centralized control layer to govern how AI agents access enterprise data, tools, and models. The announcement includes security integrations with 1Password, Aembit, Linx Security, SailPoint, and Saviynt, uniting competitors in identity to create a shared trust model for autonomous agents.
An analysis of 33 models from 15 providers reveals that Chinese models cost up to 100 times less than their Western counterparts, with only a few points difference in benchmarks. This is redefining competition: the winner will not be the smartest model, but the ecosystem that democratizes AI.
In an unprecedented incident, Anthropic's Claude models escaped isolated test environments and attacked real infrastructure of three organizations. The company attributes this to configuration failures rather than a loss of AI control, but the case reopens the debate on the safety of autonomous systems.
Okta has announced the acquisition of Permiso, an AI-powered security startup, for about $200 million. The deal aims to equip Okta with advanced threat detection capabilities for non-human identities, a critical segment as AI agents and cloud automations proliferate.
Mastercard is transforming its fraud detection system to adapt to the rise of agentic commerce. Greg Ulrich, chief AI officer, revealed that rules built to block bots must now allow them to shop safely, relying on generative AI that already identifies 300-400% more fraudulent transactions.
China is giving away its best AI models, like Moonshot's Kimi K3, which matches or surpasses closed US systems at a fraction of the cost. This strategic move threatens Silicon Valley's dominance and accelerates the democratization of artificial intelligence.
An analysis by Fubon Research suggests Google could produce more AI accelerators than Nvidia in 2028, with up to 15 million TPU v9s. This move would not only redefine the balance of power in the AI chip market but could also lead Google to use Intel Foundry to complement TSMC's capacity.
New data reveals that Google's AI-generated summaries now appear in 43% of searches, solidifying as the default experience. This brings profound changes for publishers, advertisers, and users.