A Rocket Hits the Moon and OpenAI Hides a Killer

Here is something nobody planned for: a 45-foot hunk of SpaceX hardware is going to crash into the Moon at roughly seven times the speed of sound this coming August, and there is essentially nothing anyone can do about it.

Astronomers have confirmed that the upper stage of a Falcon 9 rocket — the one that launched two lunar landers back in January 2025 — never made it to a disposal orbit and has been quietly looping around Earth and the Moon ever since. Bill Gray, who runs Project Pluto and is one of the sharpest near-Earth object trackers in the business, has pinpointed the impact at 2:44 am ET on August 5. The target: somewhere in or near the Einstein Crater on the near side of the Moon.

The numbers are almost cinematic. The stage measures 13.8 meters tall with a 3.7-meter diameter, and it will be traveling at 2.43 kilometers per second when it hits. Because the Moon has no atmosphere to slow it down or burn it up, the entire structure arrives intact. It will punch a new crater into the lunar surface and kick up a plume of debris that, unfortunately, will probably be too faint for Earth-based telescopes to catch in real time.

Gray and his team are not guessing about what this object is. They have logged over 1,000 individual observations of it since launch, tracking it through close flybys of both the Earth and Moon before its trajectory finally lined up with an impact. This is not a repeat of the embarrassing 2022 situation where astronomers initially thought a piece of space junk was a Falcon 9 upper stage, only to later identify it as a Chinese rocket body. This time, the lineage is airtight.

So why does this matter beyond being a spectacular piece of celestial trivia? Because it is a preview of a much messier future. Both NASA and China are actively planning semi-permanent outposts near the Moon's South Pole. To support those missions, the cadence of lunar launches could increase by a factor of ten over the coming decade, sending up rovers, habitats, power systems, and supply runs at a pace the space industry has never attempted before.

More rockets going up means more upper stages floating around with nowhere designated to go. The Moon's South Pole is exactly where future astronauts are expected to live and work, and nobody wants a 45-foot rocket tube arriving unannounced at 5,400 miles per hour anywhere near occupied infrastructure.

The fix is not complicated, and that is what makes the current situation mildly frustrating. With modest advance planning and a small fuel budget, launch operators can send expended upper stages into long-term heliocentric orbits — essentially parking them around the Sun in paths that keep them well clear of both Earth and the Moon indefinitely. It costs something, but it is far cheaper than the alternative of treating the lunar surface as an informal dumping ground as traffic ramps up.

For now, August 5 is circled on a lot of astronomers' calendars. The Moon will be visible across the eastern half of North America and much of South America when the impact occurs, even if the flash itself stays below the threshold of visibility. Consider it a dress rehearsal for conversations the space industry probably should have started having yesterday.

More than eight months before a gunman carried out one of the deadliest school shootings in Canadian history, OpenAI's own safety team had flagged the shooter's ChatGPT account as a credible threat of real-world gun violence — and company leadership chose not to call the police.

That decision is now at the center of seven lawsuits filed in California, brought by families of victims killed or critically injured in the attack on Tumbler Ridge, a remote mining town of about 2,000 people. The complaints paint a damning picture: trained safety experts inside OpenAI identified the account, recommended reporting it to law enforcement, and were overruled. Authorities already had a file on the shooter and had previously confiscated firearms from their home. A phone call could have placed the case back on their radar.

Instead, according to the lawsuits and whistleblower accounts reported by The Wall Street Journal, OpenAI leadership decided that the user's privacy — and the potential discomfort of a police encounter — outweighed the risk of violence. The account was deactivated. Then, almost immediately, OpenAI sent the user instructions on how to create a new account with a different email address and keep going.

Sam Altman has since called that a mistake. He traveled to Tumbler Ridge last week to deliver a public apology to grieving residents and promised that OpenAI would work with governments to prevent future tragedies. He also acknowledged that the company should have alerted law enforcement when the account was first suspended last June.

The families and their legal team are not satisfied. Jay Edelson, the attorney leading the cross-border litigation, described Altman's apology as coming far too late and offering far too little. Six of the lawsuits represent families of people killed in the shooting; a seventh was filed by the mother of a survivor still fighting for her life in intensive care.

All of the suits are being filed in California deliberately. Edelson's team wants Altman and OpenAI tried on their home turf, in front of a local jury, rather than in Canadian courts where the company had already been expected to challenge jurisdiction. That jurisdictional maneuvering, Edelson argued, was part of a broader strategy to delay any serious legal reckoning until after OpenAI completes its IPO — a transaction that recently pegged the company's valuation at $852 billion.

The timing accusation is the sharpest edge of this lawsuit. If the allegations hold up, OpenAI did not just fail to act on a threat — it actively managed information about violent users to protect its public image and its path to going public. That is a very different kind of negligence than a missed warning sign.

For an industry already navigating intense scrutiny over AI safety, this case represents a new category of liability: not what a model said, but what a company knew about how it was being used and chose to keep quiet. Whatever the legal outcome, the question of when an AI company becomes responsible for the real-world behavior of its users just got significantly harder to sidestep.