Yazılar

Italy Closes Probe Into DeepSeek After Commitments to Warn Users of AI “Hallucination” Risks

/in /tarafından

Italy’s antitrust authority has closed an investigation into Chinese artificial intelligence system DeepSeek after the company agreed to binding commitments aimed at improving warnings about the risk of AI-generated false information.

The probe, launched last June by Italy’s antitrust and consumer protection authority AGCM, focused on allegations that DeepSeek failed to adequately inform users that its AI system could generate inaccurate, misleading, or fabricated content — commonly referred to as “hallucinations.”

The decision to end the investigation was announced in the AGCM’s weekly bulletin published on Monday. According to the regulator, the commitments were submitted by Hangzhou DeepSeek Artificial Intelligence and Beijing DeepSeek Artificial Intelligence, which jointly own and operate the DeepSeek platform.

The agreed measures include clearer and more prominent disclosures explaining the risk that, based on user inputs, the AI model may produce outputs containing incorrect or invented information. The AGCM said the new disclosures are designed to be more transparent, intelligible, and immediately visible to users.

“The commitments presented by DeepSeek make disclosures about the risk of hallucinations easier, more transparent, intelligible, and immediate,” the authority said in its bulletin.

The case highlights growing regulatory scrutiny across Europe over how AI systems communicate their limitations to users, particularly as generative AI tools become more widely adopted in consumer-facing applications.

DeepSeek Researcher Voices Pessimism About AI’s Future Impact Despite Company’s Global Success

/in /tarafından

In its first major public appearance since becoming a global AI sensation, Chinese developer DeepSeek struck a surprisingly cautious tone about the technology’s long-term impact on society.

At the World Internet Conference in Wuzhen, Chen Deli, a senior researcher at DeepSeek, warned that artificial intelligence could create major social disruptions within the next two decades. “In the next 10–20 years, AI could take over the rest of work humans perform and society could face a massive challenge,” Chen said. “I’m extremely positive about the technology, but I view the impact it could have on society negatively.”

Chen shared the stage with executives from five other Chinese AI companies—Unitree, BrainCo, and others—collectively referred to as the country’s “six little dragons” of AI innovation. While praising AI’s potential in the short term, Chen stressed that companies like DeepSeek must act as “defenders” of social stability as automation accelerates.

DeepSeek rose to global prominence in January after releasing a low-cost open-source AI model that outperformed several leading U.S. systems. The company’s meteoric rise has since made it a symbol of China’s technological resilience amid intensifying competition with the United States.

Despite its success, DeepSeek has remained mostly silent publicly. Its only major appearance this year came when founder and CEO Liang Wenfeng met President Xi Jinping in February. The company has since skipped several major tech events, adding to its enigmatic reputation.

DeepSeek has continued developing its technology quietly, unveiling in September a new V3 model that it described as “experimental,” optimized for efficiency and longer text processing. Its work has also boosted China’s domestic chip ecosystem: hardware makers Cambricon and Huawei now build processors compatible with DeepSeek’s models.

In August, DeepSeek’s announcement of an upgraded model optimized for Chinese-made chips caused local semiconductor stocks to surge—underlining how the company remains both a technical pioneer and a national symbol of self-reliance in AI.

Microsoft Adds Anthropic AI Models to 365 Copilot, Expanding Beyond OpenAI

/in /tarafından

Microsoft announced on Wednesday that it is integrating Anthropic’s AI models into its Copilot assistant, marking a strategic move to diversify beyond its close partnership with OpenAI, the maker of ChatGPT.

While OpenAI’s models will continue to power Copilot by default, users will now be able to choose Anthropic’s Claude Sonnet 4 and Claude Opus 4.1 for tasks within Copilot’s “Researcher” tool and when building custom agents in Microsoft Copilot Studio.

Starting this week, users who opt in to test Claude will be able to switch seamlessly between OpenAI and Anthropic models in Researcher, said Charles Lamanna, president of Microsoft’s business and industry Copilot division.

The shift underscores Microsoft’s effort to broaden the foundation of its AI services. Until now, Copilot’s advanced features across apps such as Word and Outlook have relied primarily on OpenAI.

Although Microsoft is OpenAI’s largest investor, the company has also been developing its own models and integrating those from other AI firms. Earlier this year, it announced plans to offer models from Elon Musk’s xAI and Meta Platforms, all hosted within its data centers. Models from China’s DeepSeek have also been added to Microsoft’s Azure cloud platform.

Anthropic’s Claude models, however, are primarily hosted on Amazon Web Services (AWS), a direct competitor to Microsoft Azure, highlighting a rare cross-cloud collaboration.