Here is the most unsettling part of NASA's latest Mars finding: the carbon sitting on the surface of a Martian rock looks, chemically speaking, a lot like what you would find in ancient fossilized life on Earth. Nobody is saying Mars had life. But nobody can say it definitely did not, either.
Perseverance has been crawling around Jezero Crater for five years now, and it has found organic carbon before. The difference this time is where the carbon turned up. At a site called Bright Angel, near the edge of an ancient river channel, the rover's instruments detected complex macromolecular carbon right on the rock's surface — no drilling required. Researchers are calling it the shallowest detection of organic matter on Mars to date, which is either a genuinely exciting clue or a very elaborate red herring.
The instrument doing the heavy lifting here is SHERLOC, an ultraviolet Raman spectrometer mounted on Perseverance's robotic arm. It fires a deep-UV laser at a target and reads the light that scatters back, which lets scientists fingerprint specific molecular bonds. When SHERLOC pointed its laser at three targets at Bright Angel — poetically named Cheyava Falls, Apollo Temple, and Walhalla Glades — it returned a spectroscopic signature scientists call the G-band, indicating a dense, cross-linked network of mostly reduced carbon atoms.
On Earth, that kind of signature typically means kerogen, which is basically the compressed, fossilized remains of ancient microbes. The researchers deliberately avoided using that word in their paper. Kerogen implies biology. Macromolecular carbon just means the stuff is there, and nobody knows why.
That distinction matters more than it might sound. Mars has plenty of ways to produce carbon compounds without any help from living things — volcanic activity, meteor impacts, and certain chemical reactions can all do the job. The honest answer right now is that the data is genuinely ambiguous, and that ambiguity is itself the story.
The team also had to rule out a more mundane explanation: that SHERLOC was essentially detecting itself. Bright Angel was the first site the instrument examined after a hardware anomaly knocked out its focusing mechanism, forcing engineers to improvise a new operating mode. There was a real possibility the signal was bouncing off the instrument's own fused-silica window rather than the rock. Lab tests on spare flight hardware helped rule that out, though the team was careful about how far they pushed their confidence.
What resolves this is sample return. Perseverance has been caching rock cores for exactly this reason, and the Bright Angel site is part of that collection effort. If and when those samples make it back to Earth — a mission that has faced its own budget and scheduling headaches — scientists can run analyses that no rover instrument can replicate.
Until then, Mars keeps its secrets. The carbon is real. Whether it ever had anything to do with life is a question that might take another decade to answer, if we are lucky.
AI
Fanfiction Community Turns on Itself Over AI Detection Witch Hunt
An anonymous account on X posted a tool last week that claims to detect AI-written fanfiction by spotting hidden code left behind by Anthropic's Claude chatbot, and within days, corners of the internet that run on creative goodwill had turned into something resembling a public trial.
The tool is a custom skin for Archive of Our Own, the massive fanfiction repository with tens of millions of works. The logic behind it is technically interesting: when text generated by Claude is pasted directly from the chatbot into AO3's editor, it apparently carries along a piece of hidden formatting code called font-claude-response-body. The skin detects that code and turns the entire page background red. Simple, visual, and — in controlled tests — apparently functional.
The creator framed it as a way to protect fandom's creative culture, not to name individuals. That framing lasted about as long as you would expect. Communities quickly organized to flag specific authors whose published works triggered the red-screen response, turning a detection mechanism into a tool for public shaming.
Here is the part that should give everyone pause. The detector has a meaningful false-negative problem and a non-trivial false-positive risk. The hidden code only survives if a writer pastes text directly from Claude without touching it first. Anyone who copies the generated text into a plain-text editor first — or edits it even slightly — wipes the fingerprint clean. Sophisticated AI users would never trigger it. Only the careless ones would, which means the tool is better at catching people who are bad at covering their tracks than at measuring how much AI was actually used in a given work.
The false-positive risk is murkier but real. The code is tied to Claude specifically, and its presence does seem to require the chatbot was involved somehow. But the line between "used AI to generate this story" and "used AI to help edit a sentence" is not a line this tool can draw. Fandom's anger, however, does not tend to make fine distinctions.
This is what happens when a genuine and understandable frustration — people feel cheated when they invest emotionally in a story they later suspect was machine-generated — meets detection technology that is blunt and easy to misuse. The grievance is legitimate. Creative communities are built on the implicit promise that a human being sat down and made something, and that promise feels violated when AI enters the picture quietly.
But weaponizing an imprecise tool against specific individuals in public forums is a different thing entirely from setting community standards. Writers are already reporting anxiety about whether their work will be flagged, regardless of how they actually wrote it.
Anthropic has not commented on whether the detector works as described. The broader question — how creative communities establish trust in an era when AI can imitate human writing well enough to fool most readers — is one that a browser skin cannot answer.
Source: The Verge
The tool is a custom skin for Archive of Our Own, the massive fanfiction repository with tens of millions of works. The logic behind it is technically interesting: when text generated by Claude is pasted directly from the chatbot into AO3's editor, it apparently carries along a piece of hidden formatting code called font-claude-response-body. The skin detects that code and turns the entire page background red. Simple, visual, and — in controlled tests — apparently functional.
The creator framed it as a way to protect fandom's creative culture, not to name individuals. That framing lasted about as long as you would expect. Communities quickly organized to flag specific authors whose published works triggered the red-screen response, turning a detection mechanism into a tool for public shaming.
Here is the part that should give everyone pause. The detector has a meaningful false-negative problem and a non-trivial false-positive risk. The hidden code only survives if a writer pastes text directly from Claude without touching it first. Anyone who copies the generated text into a plain-text editor first — or edits it even slightly — wipes the fingerprint clean. Sophisticated AI users would never trigger it. Only the careless ones would, which means the tool is better at catching people who are bad at covering their tracks than at measuring how much AI was actually used in a given work.
The false-positive risk is murkier but real. The code is tied to Claude specifically, and its presence does seem to require the chatbot was involved somehow. But the line between "used AI to generate this story" and "used AI to help edit a sentence" is not a line this tool can draw. Fandom's anger, however, does not tend to make fine distinctions.
This is what happens when a genuine and understandable frustration — people feel cheated when they invest emotionally in a story they later suspect was machine-generated — meets detection technology that is blunt and easy to misuse. The grievance is legitimate. Creative communities are built on the implicit promise that a human being sat down and made something, and that promise feels violated when AI enters the picture quietly.
But weaponizing an imprecise tool against specific individuals in public forums is a different thing entirely from setting community standards. Writers are already reporting anxiety about whether their work will be flagged, regardless of how they actually wrote it.
Anthropic has not commented on whether the detector works as described. The broader question — how creative communities establish trust in an era when AI can imitate human writing well enough to fool most readers — is one that a browser skin cannot answer.
Enjoyed this?
Get stories like this delivered every Tuesday — free.