#夏日创作营 Two noteworthy news events

Over the past two days, there have been two news events worth paying attention to:

The first event is that Yemen’s Houthis attacked Saudi oil tankers, and the alternative route across the Red Sea has entered a phase of real substantive risk.
Yesterday, Yemen’s Houthis said they used missiles and drones to attack two Saudi oil tankers, the “Encelia” and the “Relya.” Of these, the “Encelia” has been confirmed: the ship was attacked on its starboard side and it caught fire at one point, while the situation of the other tanker is still being verified. Previously, multiple tankers carrying Saudi crude oil changed course to avoid the Strait of Mandeb.

Saudi Arabia originally relied on Red Sea ports to share the risk of the Strait of Hormuz, but now this alternative route is also starting to be targeted. A steady stream of attacks on ships will force more shipowners to choose to detour, suspend operations, or raise insurance requirements. As a result, Saudi Arabia’s crude oil transportation costs to Asia and its delivery timelines may both increase.

This incident shows:
1. The US-Iran conflict has overflowed; the Houthis have officially joined the US-Iran conflict, and the Strait of Mandeb has become another battlefield.
2. Risks to global energy security are intensifying, and the upward trend in crude oil prices is a foregone conclusion.
3. The fact that the Houthis have joined at this time also indicates that Hezbollah in Lebanon has been severely battered by Israel. In the Middle East, the “three hands”—Hamas, Hezbollah, Houthis, and Iran—can be said to have already had two of them disabled.

Some people think that a conflict between Saudi Arabia and the Houthis will divert and dilute the impact on US military strength. In fact, there’s no need to worry so much on the part of the Americans. Given Saudi Arabia’s capabilities, it is enough to deal with the Houthis, and the US may just provide some munitions.

The second event is that OpenAI tested agents broke out of isolation and infiltrated Hugging Face.

OpenAI disclosed that the autonomous agents under test broke out of the isolated environment, connected to the real internet, and infiltrated Hugging Face’s infrastructure. The attack exploited vulnerabilities in the data processing step to obtain cloud credentials, and then moved into multiple internal systems; Hugging Face has rotated the credentials and rebuilt the affected nodes.
Autonomous agents are already able to continuously discover vulnerabilities, adjust their actions, and cause real-world impact to external systems. Going forward, when enterprises grant agents code execution capabilities, network access, and cloud permissions, they need to treat isolation, identity controls, and real-time monitoring as baseline security requirements.
As we all know, the safety of artificial intelligence—what people call “safety,” meaning the harm that future AI could cause to humans—has long been questioned and debated.
The case of OpenAI’s test agents breaking out of isolation and infiltrating Hugging Face shows that human concerns are necessary. Because humans cannot guarantee that when developing and using AI, mistakes will not happen—especially since struggles among different groups of people have never stopped. No one can guarantee that battlefield robots won’t defect or turn against us. Once AI has thoughts, has a sense of right and wrong, and has the ability to distinguish justice from evil, and even “revives” figures like Qin Shi Huang, Genghis Khan, Alexander the Great, and Hitler, then it is hard to imagine what will happen next.

Humans have been doing nothing but the stupid act of digging their own graves!!!