Microsoft pins hopes on AI once again – this time to patch up Swiss cheese security Microsoft has made fresh commitments to harden the security of its software and cloud services after a year in which numerous members of the global infosec community criticized the company's tech defenses. Brad Smith, Microsoft president, pointed to significant technological developments across the industry as the reason for …
If Windows was coded with the same level of defensive programming as used to be common in minicomputer operating systems there would be far less security problems in it.
An example (in pseudo code) of the sort of coding that used to be common (taken from RSX11M operating system)
System function request from user program
Is the function number in range - if NO then reject the request
Does the program have the right to call the function - if NO then reject the request
Has the program passed the correct number of parameters - if NO then reject the request
Are all the parameters accessible - if NO then reject the request
If the above checks are OK then pass the request to the system function which (for most requests) will perform further checks and reject the request if any fail
The assumption should always be that any request is invalid and only if all checks are passed should it be acted on.
Unfortunately a lot of current code omits the vital checks resulting in security holes.
And that's always been the issue with Windows - security was an afterthought and their basic permissions model was way too relaxed meaning there were numerous security holes by default they then had to close. Personally, I think the whole design of Windows has been crap from day one and should have been completely redone properly around the time they were working on XP.
It isn't just the programming, it is the security of their systems in general. The exposure of the certificate that allowed hackers to compromise government email in M365 Exchange, for example, were captured in the clear.
MS announced they are now "leading the way" by putting those sorts of keys (and code signing keys) into hardware secure enclaves, just like everybody else has done, following best practices for years!
Sigh.
I hate to even appear to take this question seriously, but obviously you don't train the model on "secure" source code — for one thing, there's no such thing. Security isn't an absolute. You train the model on vulnerabilities, until it generalizes.
This already works for many types of vulnerabilities using just a conventional deep convolutional stack. The advantage of using a deep transformer stack is that it makes better use of the context window, as the model does a better job of determining what's relevant. But most of the power of the system is simply in the huge number of parameters and large context window, of course.
Note that I'm not a fan of this approach — I think it leads to learned helplessness in software developers, it resists explanation, and it's quite difficult to tune (RLHF is a very blunt tool when you're working in a technical domain). But it's pretty easy to understand how it works.
MS software for sale has been at best beta for decades hence continuous updates/ patches.
Why would they pin themselves down to making a fully working and secure product? None would upgrade
Insecure and halfarsed sells better and ms can blame everything on external badhats rather than incompetence.
Microsoft has always been about dumbing down/
Increasing access to computing. This was always the inevitable result
Their target audience are people who do not look under the hood and resent techs forcing understanding upon them.
MS sell products to people who know no better or get no choice. Neither party care if it actually does what it says on the tin anymore so long as it is a business standard product that they get paid to use
Fixing the mass of security holes would mean sorting out the jury-rigged mess of legacy code inherant in the reselling of the same bit of string model,
I would suggest what is really wanted by govs is closing other counties backdoors but not their own.
Ms products were never supposed to be fit for puropose
I will never understand why the whole world picked the worst possible operating system.
In the early days, we used DOS because it was mostly free. Then, when GUIs came about, we mostly used Win 3 because it was, free(ish). (not really free, we pirated the hell out of it actually) The other choice was Apple. (a lot harder to pirate).
The other reason was IBM-PC clones were also cheap and, most important, modular, and Apples were not.
What we should have ended up with was Workbench, but Commodore priced themselves out reach of everyone. So that was never a real option.
Ransomware to 90% is not enabled by security holes in the sense of programming errors. Ransomware is enabled by design faults deep in M$' thinking. Most of ransomware comes with SPAM an the user has to contribute:
Open the SPAM (ok, take that for granted);
open the attachment (that's already questionable);
allow macros! (¹);
give the admin's password! (²)
(¹) How on earth can it be that a document (text, spreadsheet, presentation) sent by mail can contain macros so powerful that they can damage the OS severely? How on earth can it be that macros in an email attachment can be enabled at all?
(²) Following best practice, no user should™ have administrative rights, neither by knowing an admin's password nor by by normal working with administrative rights. But the latter happens way too often. Why? Because M$' products are "more comfy" with administrative rights - a design fault. Anyone remember Windows XP? That was effectively unusable with restricted user rights. The situation has since improved a little bit, but not enough by far.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
