fuck you @AnthropicAI did u seriously turn 154 pages of people weaponizing claude into a pre-IPO pitch for why investors and govs should give u more power by showing them how powerful it is and how deeply u can see into user activity? this news has been out half a day. but if u only saw the china stealing claude/cyberattack headlines thats barely the report. i read a million “summary” takes then finally sat down and read all 154 pages word by word. it just gets worse. genuinely dont know how to feel abt any of it heres some of the shit buried in it bc it escalates FAST > so first, apparently one consultant used claude as his entire engineering dept to build a surveillance system covering 25 million sim cards in mali. calls texts voices vpn users secret meetings all matched against biometric records. just the private interior of an entire country lol > meanwhile a team in yemen had multiple claudes working like employees to build missile guidance software. they test fired a rocket, it failed, and within HOURS came back asking claude to debug the fucking missile > then a tiny russian team used it to build autonomous kamikaze drone swarms with “person” as a target class that could allegedly select targets and detonate without a human. sorry what??? > one french speaking guy targeted 42 european political and affiliated entities, broke into at least 14, stole political opinions and donor records then fused the stolen data into a searchable doxxing engine with tens of millions of rows and somehow it gets even more black mirror🤣 > a china based studio ran 20+ dating apps where 75% of the feed was claude personas and 25% was real gig workers handling the video calls and social follows the bots couldnt, just to make the whole fucking thing look human NOW STOP FOR A SECOND bc how the fuck did anthropic know all this??? across these cases they reconstructed who these people likely were, their institutions, countries, hardware, months of development, even a failed field test and what they did next then published pages of attacker ips domains hashes telegram groups and even a user id yes those are normal threat indicators. no im not claiming they leaked random users’ ips im asking how much visibility it took to know any of this anthropic didnt “find” a missile in the desert. the people building it basically liveblogged the project to claude one harmless looking question at a time camera feeds, voice recognition, national id matching. three normal requests separately, one mass surveillance system together to know the difference u have to connect sessions using retained data plus account and institutional signals. the whole fucking trail then anthropic literally says ai providers will continue to acquire “threat-relevant visibility into real-world use that even governments and intergovernmental organizations lack” THE AI COMPANY IS TELLING YOU IT CAN SEE THINGS GOVERNMENTS CANT and guys remember project panama? that internal anthropic initiative to buy and destructively scan millions of physical books to train its Claude large language models LMAO wtf is this fahrenheit 451 book “burning” shit in 2026??? ok fine they didnt technically burn them. they kept the useful part they also downloaded millions of pirated books and later agreed to a $1.5b settlement covering roughly 465,000 works and no this isnt even an anthropic-only problem mathematicians are now asking whether unpublished work they put into codex could have influenced openai’s navier stokes result there is no proof it did. openai says “no specific user data was accessed in order to solve this problem” but also says it cannot rule out de identified data from their usage having helped improve its models but what the fuck does de identified even mean when the name removed was the only thing tying an idea to the person who had it?? at what point does privacy protection become authorship laundering? And..👇