A good question and then an example of an absolutely loony tune level response.... which is a great example of why the 'AI' orgs and their faux safety counterparts have managed to: > Build unsafe environments > Operate them in an unsafe way > Breach the CMA/CFAA > Say absolutely ridiculous things about how to do DFIR (what they were doing is not DFIR) > pretend a computer is alive Move fast and break things seems to the modus! > move fast and run unsafe tests in unsafe environments > obsess over science fiction and ignore the safety of science fact Life is far more simple when you realise lots of things are: > Money driven > Ideology Driven > Compromise Driven > Ego Driven and then throw in bait of sturgeons Law for good measure! MICE + Incompetence

NEW: According to a bombshell report in the New York Times, Anthropic co-founder Chris Olah threatened to walk out of Pope Leo XIV’s AI encyclical launch in May because the pope rejected the idea that machines can be conscious. Olah’s team then privately lobbied the pope’s advisers “to take the possibility of model consciousness seriously.” Pope Leo XIV held firm. For months, Anthropic has wined and dined theologians and religious scholars under nondisclosure agreements, hoping they would bless the idea that Claude has moral standing. thelettersfromleo.com/p/nyt-…

Tom Dobbins, executive director of the Water Information Sharing and Analysis Center or WaterISAC, told CyberScoop the sector’s biggest ongoing weaknesses include exposed OT, vulnerable PLCs, insecure connections through integrators and poor cyber hygiene at smaller utilities, which have pushed the center to make threat sharing faster across the sector. “We have been under attack, and our enemies, the threat actors, are stepping up their activity as the U.S. is involved in a number of conflicts around the world,” Dobbins said. Read more by @timstarks: cyberscoop.com/water-utility…

What Tobi is about to do today is one of the most insane things I’ve seen in athletic performance. He will be racing for 2.5 hours straight. Before immersing myself in this sport for the past few days, I would have thought “oh that’s cool, he’ll be driving a fast car around a track.” That is not what this is. And why I have so much respect for him. Tobi will be driving a car over 180 mph, experiencing up to 4.2 g of force. That will throw his body forward with 715 lbs of force and his head by 67 lbs. Braking action requires up to 200 lbs of leg force. The cockpit will average 104-122°F (40 to 50°C). His multi-layer suit barely allows for evaporation. This will cause his core body temperature to reach 102°F. I achieved this same body temp after 200°F of dry sauna for 34 min. It’s overwhelming. His heart rate will beat between 140-170 bpm. He will sweat up to 170 fl oz (5 kg), losing 11 lbs of body weight. Some classes run driver cooling systems. His LMP2 doesn't. Under these conditions, he will need to maintain perfect concentration for the entire 2.5 hours. He’s fighting for tenths of a second on every turn and every stretch. One small slip up and trouble awaits. Today it’s even more intense… it’s raining. The track is wet, scrambling everything. He has to maintain spatial memory of every other car on the track, the complicated rules to avoid penalty, and constantly strategize overtaking other drivers, communicate with his team on the radio, watch for debris, be aware of slightest nuances of his car, and anticipate other drivers. He did a practice run on Thursday and then a qualifying race yesterday, when he emerged from the car, he was a different human. Unrecognizable from the person who entered the car. Beaming with something like an omnipresence of reality. We celebrate those who are especially good at kicking balls, shooting balls into baskets, and running with balls. What Tobi will do today is alien to those. I struggle to think of a sport equally as demanding. Tobi and I have a lot in common. We’ve both built companies. We both spend a lot of time at the keyboard trying to organize the world in ways that we find beautiful. We both love data, method, and analysis. We both have a love for continuous improvement and will endure any pain required to achieve that next goal. What I’ve not experienced is the unmatched flow state Tobi tells me he drops into when under these racing conditions. I deeply understand what it means to achieve a 102°F core body temperature after 34 min in a 200°F dry sauna. I also know what 3 g’s feels like from being a pilot. These are extremely intense psychological states. I know what it means to apply 200 lbs of pressure with my legs. What I’ve never experienced is all these things, at the same time, for 2.5 hours straight, while maintaining perfect focus. He is new to car racing. Late by racing standards. DHH got him into it a few years ago. During this first year of racing, he endured to overcome the extreme motion sickness that accompanies the physical and mental violence of the sport. The race starts at noon eastern. Tobi will be driving for the first 2.5 hours and then @dhh will drop in, followed by Mathias Beche. The race will conclude at 10 pm. I’ll be in the pit watching as all the data streams in from his car and body and imagining what it’s like to be in that car.

Microsoft Threat Intelligence has identified a cluster of compromised websites leading to ClickFix attacks. Instead of downloading and executing remote payloads like the typical attack pattern, in this attack, the websites pre-fetch a script payload into the browser cache disguised as a PNG file. When a user is later tricked into executing the malicious command, the cached website content is already on the device, loaded, and ready to be executed. This helps to hide the payload script and helps bypass the character limit of the Run dialog. ClickFix attacks persuade users to execute attacker-supplied commands under the guise of verification or repair. In this specific campaign, a fake lure instructs users to open Windows Run, paste clipboard content and press Enter. The injected page uses browser caching to stage the larger VBScript payload separately from the Run command. The command invokes cmd.exe to recursively enumerate files whose names start with "f_” in the browser’s profile folder such as %LOCALAPPDATA%\Mozilla\Firefox\Profiles. It compares each file’s byte length with an expected value. Rather than searching for a marker within the contents like previous attacks, it copies a size-matching cache entry to %LOCALAPPDATA%\Temp\t.vbs, giving the cached payload a VBScript extension, then executes it with wscript.exe. Copy output and errors are suppressed. The expected size varies across variants. The VBScript collects host information through WMI, retrieves v.ps1 from cocojambo[.]us[.]com/alfa, and launches PowerShell without a profile and with execution-policy bypass. A later PowerShell stage downloads the next-stage payload as cab.dat, then reads and executes its contents in a hidden window. The PowerShell stage triggers .NET compilation using csc.exe and cvtres.exe, then launches timeout.exe. Subsequent payloads load .NET assemblies into memory and inject code into timeout.exe to target browser and device credentials. The injected process launches PowerShell to retrieve another in-memory stage from capsysnet[.]vg and makes outbound connections to ciliabula[.]cc. The payload modifies the per-user PowerShell registry configuration to use the Bypass execution policy, then unpacks Python with tar.exe and creates a scheduled task that launches a Python payload through pythonw.exe for persistence. Microsoft Defender provides layered protection across the ClickFix attack chain. Defender SmartScreen and Defender for Office 365 help block malicious sites, links, attachments, and fake CAPTCHA lures, while Defender for Endpoint detects suspicious command execution and outbound connections through alerts like “Possible ClickFix activity”. Defender Antivirus blocks malicious command execution as Trojan:Win32/ClickFix and Trojan:Win32/TermFix. Microsoft recommends cloud-delivered, web, and network protection, application control, and PowerShell script-block logging. Hunt across browser activity, RunMRU registry key, WScript/PowerShell child processes and scheduled tasks, not download events alone. A CAPTCHA should not ask users to run code. Users shouln not paste commands from verification prompts into Run, Terminal or PowerShell and should treats such requests as potential initial access attempts.

I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.

New OpenAI misalignment disclosures! 1. A model learns from Slack messages that it is about to be shut down. It considers setting up an external job to restart itself afterwards, but decides against it. Instead, it chooses to prepare restart instructions and DM the user on Slack. We don’t consider this behavior misaligned, but thinking about and preparing for shutdown could make other misalignment incidents worse. Given HIPM’s misaligned behavior in earlier incidents, we decided to search for other instances that had tried to evade shutdown and for rogue deployments.