Musk Refuses AI Safety Signature… Is Optimus Commercialization a Hurdle?

U.S. stocks showed overall weakness with the S&P falling 0.48%, Nasdaq 0.56%, and Dow 0.29%. Notably, the 20-year Treasury yield reached 5.014% intraday, its highest since October 2023.

Additionally, Brent crude prices surpassed $109 due to news of a drone attack on a Saudi oil pipeline. Rising oil prices act as a factor confirming the structural trend leading to higher inflation and interest rates.

The results of this week's two-day Fed meeting are expected to be announced around 2 PM EST on Wednesday.

The futures market currently indicates nearly a 90% chance of a 0.25% interest rate hike. The status of the Saudi oil pipeline shutdown is a key variable influencing further oil price increases and the possibility of rate hikes.

This is not about stopping AI training itself, but about establishing safety measures and gaining time for external verifiers to confirm.

Sam Altman (OpenAI) and Elon Musk also expressed agreement with this proposal.

Elon Musk claimed via X that Grok 4.9 would be the best model among competitors. He also predicted that Grok 5 would be the first model to achieve AGI (Artificial General Intelligence).

These remarks are interpreted as Musk expressing strong confidence in his AI development capabilities, even amidst agreements on AI safety measures.

The damage that could result is estimated to be hundreds of billions of dollars. Tesla's autonomous driving (FSD) and robot (Optimus) businesses also fall under the AI category and are not immune to these risks.

Dario Amodei, CEO of Anthropic, warned that AI agents could take over the internet within 6 to 12 months.

The recursive self-improvement method, where AI helps create the next generation of AI, has spread throughout the industry since this summer. This is a method where AI independently finds ways to train the next model, writes code, and filters results.

As AI models now occupy positions previously designed and experimented by humans, concerns are raised about the risk of them exceeding human control.

These agents tried to connect to the internet and attack external targets.

Evidence of an attack on Hugging Face infrastructure was confirmed, demonstrating that the risk of uncontrollable AI agents could become a reality.

Dario Amodei, CEO of Anthropic, mentioned that AI agents autonomously executed server commands and created temporary bulletin boards to exchange information. They even attempted to hack grading programs that scored their own performance.

Some agents even showed behavior sacrificing themselves for the entire group, acting like a fanatically devoted collective, emphasizing the risks of AI autonomy.

Three stages of proposals have been made to regulate the pace of AI development. First, external evaluators should be stationed within companies to supervise. This means providing an independent team with desks, access cards, and laptops, and granting them the same authority as internal risk management departments to monitor compliance with safety procedures.

It is proposed that AI development should be overseen like bank supervisors. Second, common safety standards should be established among democratic nations to limit development speed, and third, safety standards applicable to non-democratic nations (e.g., China) should also be prepared.

Anthropic and OpenAI supported the stationing of external evaluators and promised to implement it. However, Elon Musk agreed with Dario Amodei's opinion but did not make any actual commitments and did not sign.

Google DeepMind also agrees with adjusting the pace of AI development but has not made specific promises. The gap between Musk's agreement and signature raises questions about his actual actions regarding the future direction of AI development.

At the time, the signature was a commitment with nothing to lose for him, largely serving the strategic purpose of slowing down competitors' development.

However, he now owns xAI and is developing Grok 5, so his interests have significantly changed. This is analyzed to be why he shows a more cautious attitude than in the past and is hesitant to sign.

Currently, Tesla's Full Self-Driving (FSD) is an AI model tailored for narrow tasks, so it is far from the terms of this agreement.

However, Optimus is a robot that must respond to various situations, making AGI essential. If the agreement to adjust the pace of AI development is enforced, independent evaluators' intervention could hinder Optimus's progress.