That is what kept creating in my mind, it was pushed there by the BBC (at https://www.bbc.com/news/articles/cqkg50j1yd5lo) less then 24 hours ago and it is time that I (kinda) repeat what I have been saying all along and for this version lets use the ‘fairy tale’ approach. The title gives us ‘Rogue Anthropic AI agent gave police fake tip in unsolved murder case’ and I get (to some degree) fed up with this rogue setting. So lets explain the setting of a rogue.
This is the first rogue, in this case a female rogue and thoroughly explored in Dungeons and Dragons themes where to poor butcher (could be a fisherman) pleads with this rogue “Oh, please gloomy dark lady, do not deprive me of my virginity” and the outcome of that is decided by dice (or an interrupt card) and that is the ‘Rogue’ that has been in play for centuries (although until the 18th century I have never seen a female version. I did hear from a man in the 19th century named Dick Turpin that he met one, but it might have been the combinations of ale and wines and him losing his purse.
The decently real rogue in this case is the ‘dorky’ IT guy who gets his programming a little wrong. Like the ‘eager’ student in a renowned University working on his ‘assignment’ and he is looking into statistics, more precise Multivariate Analysis of Variance (aka MANOVA) and the dastardly scoundrel gets a parentheses wrong the “(“ or “)” particularly the placement and now what was a small assignment where the illustrious student evaluates a linear combination of multiple correlated outcome variables, but as one of the parentheses is in the wrong place, the mainframe it working on is seeing a correct statement and it does it, but the size of what is to be analyzed take a lot more than the available memory of the mainframe. It goes a little bonkers. And makes the system administrator scream and call for his mommy.
And as the BBC is giving us “An artificial intelligence (AI) agent developed by Anthropic went rogue and sent US police a fake tip about an unsolved murder earlier this year, authorities have revealed. The Philadelphia Police Department said the 18 July tip was “flagged as spam” and not passed on for investigation – but the force criticised the tech firm for taking more than two months to detect and report the breach. It said the bogus tip came through a public website where people can share information on unsolved murders, and that the AI agent had claimed to have seen “someone matching the description”.” So there are clams that could be it, but this is not some system going ‘rogue’ because that system is not armed with sharp blades (or poisonous arrows) and the setting that the BBC is giving us (and many other media as well) is a programming error like I described, but these systems cannot afford the class actions coming from there, as such we now get more and more ‘class actions’ Now this last part is speculation by me. I would call it presumption, but that requires genuine insight and I don’t have it, because to me all AI is fake. But consider that predictive analytics uses statistics to find causality and correlation and optionally ANOVA and MANOVA are used. Now ask yourself could a programmer gotten a parentheses wrong, of will we see servers stealth fully walk the corridors of your corporation, leaving false evidence behind?
Just something to consider.
So whilst we get “The police department added that there were no signs of breaches to any departmental systems and that its safeguarding processes stopped the fake tip from getting past its spam folder. But the safeguards “do not diminish the seriousness of an AI system presenting fabricated information as though it came from a person with knowledge of a homicide”, the force said.” And there is a case for “AI system presenting fabricated information”, in courts there have ben several cases where these systems ‘created’ false case files and not once, but several times. And in the age of utter shortcomings in validation and verification, these systems are extremely flawed. I wonder how these systems hold up against the system that they created, The UK Police Scientific Development Branch started the Major Incident Project in 1974. It would be called Holmes and apparently (I was unaware) a Holmes 2 system was rolled out around 2000. So, were there considerations on how validation and verification happened in those days?
And when we get to “Anthropic this week published a report detailing multiple types of “unintended” actions its agents have taken. Organisations that had been impacted also included several US government agencies including the White House, it said. The US State Department said the AI agent had filed 20 visa applications using a form on its website, but that they were incomplete and not processed, according to reports. US President Donald Trump recently announced an AI taskforce, which he said will co-ordinate engagement between the government and all parties, including AI companies, consumers and religious groups.” So in this case was the agent a real person? And when we get to “the safeguards “do not diminish the seriousness of an AI system presenting fabricated information as though it came from a person with knowledge of a homicide”” was this a person wanting (read: hoping) for a free ride into the United States? The article left me with questions, as well as the fact that “presenting fabricated information” requires certain steps of validation, optionally veridic action and at the very least question into the predictive analytics side of things, because a programmer getting (optionally) a parentheses wrong is seemingly one with enormous potential and optionally the easiest solution into this setting. It is not a setting that Anthropic likes, because that could start a new class action war and when one goes down the others go too and I reckon that OpenAI, Grok, and Gemini will assist in keeping the ‘Rogue’ setting alive, they all have plenty to lose and my giggling sense of humour comes with the setting that these fake AI settings is costing them 3-4 trillion and the court cases settlements could exceed that at the very least (but that is merely my sense of humour)
So have a great day and consider when you get your coffee that the POS could steal your credit card, Rogues are intensely able to do this and spend it on new shoes. Or we could look at where the programmer went wrong. Its all up to you.
Have a great day
