Anthropic and OpenAI Discuss AI Ethics with Religious Leaders

Representatives from Anthropic and OpenAI joined the first "Faith-AI Covenant" roundtable, held in New York, to exchange views on AI ethics with leaders from multiple faiths.

The move by AI companies to seek ethical guidance from religious communities marks an approach quite different from the technology-driven discussions of the past. Critics, including AI researcher Rumman Chowdhury, have dismissed such dialogue as "at best a distraction," warning that it risks diverting attention from concrete debates over regulation and control of AI systems.

Bringing in religious and philosophical perspectives is meaningful in itself as a reflection of diverse values, but the key question going forward will be how this translates into actual AI regulation and safety measures.

ByteDance Raises AI Investment to Over $30 Billion Amid Shift to Chinese Chips

ByteDance, the parent company of TikTok, has raised its 2026 AI-related investment plan to more than 200 billion yuan (roughly $30 billion) — an increase of at least 25% over its previous plan.

What stands out is that the company is reducing its reliance on U.S.-made chips and accelerating its shift toward Chinese semiconductors. Still, compared with the combined $725 billion that Google, Amazon, Microsoft, and Meta plan to invest, the gap in scale is stark.

For Chinese AI companies, U.S. semiconductor export restrictions are a major constraint, and the switch to Chinese-made chips is less a strategic choice than a near inevitability. Even so, a $30 billion investment demonstrates that ByteDance is serious about competing in AI.

METR Reports Measurement Limits of Claude Mythos, Warns of Autonomous AI Attacks

METR, an AI evaluation organization, has reported that Anthropic's Claude Mythos Preview is "nearly unmeasurable" with current test suites. Of 228 test tasks, only five cover the relevant range of capabilities.

This highlights a fundamental problem: model capabilities are outpacing the evolution of evaluation methods. The argument that "lagging evaluation methods," rather than the models themselves, are becoming the bigger challenge is an important point of debate in AI safety.

At the same time, Palo Alto Networks has reported that frontier AI models can autonomously chain vulnerabilities, completing an attack from initial access to data exfiltration in as little as 25 minutes. As a concrete sign that AI-driven cyberattack capabilities are approaching practical use, it warrants vigilance.

Forbes: AI Startup Offers Software That Monitors Employees' Work

According to Forbes, an AI startup provides software that monitors employees' work in real time and promises to reduce "inefficiency."

AI-powered employee monitoring tools aim to boost productivity, but they also raise concerns about privacy and working conditions. Securing worker consent and transparency is likely to prove contentious.

Since the pandemic, demand for employee monitoring technology has grown alongside the spread of remote work, but it is worth noting that AI is dramatically expanding both the granularity and the scope of surveillance.

Draft Transparency Rules for EU AI Act Article 50 Published

The European Commission has released draft guidelines on implementing the transparency obligations under Article 50 of the EU AI Act. The guidelines define how the transparency requirements for certain AI systems should be put into practice.

On Hacker News, commenters have raised concerns that these transparency rules could produce a mess similar to the one that surrounded cookie consent implementation. Avoiding a repeat of the "checkbox culture" — where users are merely asked for perfunctory consent — will test the specificity and effectiveness of the guidelines.

---