NHS to close-source hundreds of GitHub repos over AI, security concerns

The UK's National Health Service (NHS) is ordering all of its technology leaders to temporarily wall off the organization's open source projects over concerns relating to advanced AI and Anthropic's Mythos. According to guidance shared internally within the organization and seen by The Register, GitHub repositories must be set …

[Brewster's Angle Grinder]

I take it this will be one of those permanent temporary closures - where a cash-strapped public sector organisation never quite has the resources to ensure it is safe for public release and so it remain temporarily closed for ever?

[elsergiovolador]

"Nothing is more permanent than a temporary fix"

[Fara82Light]

The NHS is not cash-strapped.

[SnailFerrous]

A mixture of security by obscurity and bolting the stable door after the horse has departed. Neither are known for their success rate.

[AndersH]

Nothing endures like the temporary.

This has got to be a decision taken by someone senior with limited technical knowledge.

[elsergiovolador]

Re: Nothing endures like the temporary.

More like looking for an excuse to hide poor code quality delivered by usual suspects.

[Innominate Chicken]

Re: Nothing endures like the temporary.

Starts with a C, continues with a music genre, then a comment icon, and finishes with an A.

[elsergiovolador]

Re: Nothing endures like the temporary.

I smell poo.

[Anonymous Coward]

Re: Nothing endures like the temporary.

Wes Streeting?

[elsergiovolador]

It works for Apple very well. People still believe iOS is super secure, thanks to the fact they cannot run any anti-virus on it.

[Fustbariclation]

Insecurity through obscurity

It's worse than that. Making it closed-source will give a dangerously false sense of security.

Existing open source code will still be useful to bad actors.

Besides, with the old source code as a guide, even fairly pedestrian LLMs, with a bit of encouragment, sould be able to reverse assemble the new closed-source binaries to reveal updates in plain code.

Newer LLMS should make short work of breaking code hidden in this way.

Meanwhile, they miss the many-eyes security protection that Open Source can give.

[PenfoldUK]

Sounds like an excuse to close source the NHS.

How are firms supposed to make billions off the back of the NHS if code is open source?

I doubt it will prevent threat actors exploiting weaknesses. But it will prevent AIs and prople detecting and patching them.

[elsergiovolador]

Not only that. The usual suspects had to maintain the notion of quality in case Joe Public starts digging through the repos. Now that it is hidden, it will become spaghetti code galore - my guess.

[steelpillow]

H'mm

I'm guessing it's a measure to cut the flood of AI vuln reports, so that ransom gangs are less able to identify and exploit them before they can be patched.

But hey, this is Socio-Political Bureaucracy, and in the UK to boot, so I'm not placing any bets.

And is GitHub secure enough to deliver anyway? Now who owns it, can anybody remind me?

[Anonymous Coward]

Re: H'mm

Using other people’s code or libraries without any idea at all what they contain or do - always a security winner for lazy programmers.

[An_Old_Dog]

Public Repos are *PUBLIC*

"Public repositories materially increase the risk of unintended disclosure of source code, architectural decisions, configuration detail, and contextual information..."

Disclosure is the reason for making repos public. It's an intentional act.

This sounds like an attempt to retroactively compensate for poor development practices (putting credentials in repos, leaving [at management's orders] security "for later", etc.).

[heyrick]

"We will continue to publish source code where there is a clear need."

And there will never again be "a clear need".

How about this. If it isn't being done in the public good (publishing source), then don't expect public money to pay for the development any more. You can take it out of management budgets...

[Graham Cobb]

Have they checked their licences?

Have they checked the licences of software they rely on?

IANAL but I would have thought copyleft licences might require software under those licences which they have modified and then distributed to be published. What counts as "distribution"? I don't know.

Even permissive licences may require attribution to be published somewhere. Particularly where some sort of "service" is being provided using the software.

I'm not saying everything has to be published, but that unpublishing definitely requires a legal review before it happens.

[doublelayer]

Re: Have they checked their licences?

Pretty much nothing requires that except AGPL, and then only to users which could be done without having a repository on GitHub. Open source licenses don't mandate how and where you distribute the sources, at least not this specifically, and most of them won't apply to most of this code.

[geospacedman]

Re: Have they checked their licences?

Once something is open licensed, that version is out there, it can't be "unpublished", and your licensed rights can't be revoked. People are already panic-archiving NHS repositories and duplicating the current versions in anticipation of a mass `rm -rf /github/nhs` event.

[Anonymous Coward]

Re: Have they checked their licences?

Do it, the license explicitly allows it - though you only have until May 11th

[Anonymous Coward]

Re: Have they checked their licences?

MIT is the default license, so you are free to fork all the projects right now

[Cloudseer]

Re: Have they checked their licences?

Default for published software is all rights reserved, unless NHS have standardised on MIT?

[VoiceOfTruth]

Another step towards privatisation

But the NHS has no problem granting Palantir access. Hmmm.

[Anonymous Coward]

Re: Another step towards privatisation

Access to "the reptile" but for us.

[Roland6]

Re: Another step towards privatisation

Or leaving the code on GitHub for Copilot to harvest…

[Gavsky]

I suppose it makes some feel a bit safer, but as mentioned in the article: bolted, door, closing, horse, stable (I paraphrase, ahem). Government & public bodies are masters of "Oh, is that a potential risk, then? Erm, somebody should really do something about that"...

[BartyFartsLast]

And then the kudos for "being seen to be doing something"

[Anonymous Coward]

correct horse battery stable?

[mpi]

And that is supposed to accomplish...what exactly?

It's a pretty safe bet to assume that any repo, that was public at any public git/DevOps platform at ANY point in time, has been scanned, forked, cloned, or otherwise read by at least one automated system somewhere, and in all likelyhood more than once. Meaning, a complete copy of any such code is almost guaranteed to already exists somewhere else.

So, what good is closing the repos now?

[Ken Hagan]

Re: And that is supposed to accomplish...what exactly?

As the original article points out in it's closing paragraphs.

But I think you'll find that the purpose of today's move is more about creating the appearance of doing something than it is about meaningful security.

[b1k3rdude]

All of said projects should never have been public facing from the get go.

[Ken Hagan]

False positives

"Anthropic has still not yet revealed the number of false positives the model throws up when running vulnerability scans, which is a common issue with AI thus far."

Neither has it made much noise about how many could be found using traditional tools like compiler warnings, lint, and valgrind, all of which are much cheaper than AI.

[Anonymous Coward]

Re: False positives

But not if someone else runs them against you...

[WSWS]

Re: False positives

If you're scanning existing code that has been around for a while, you can assume that those traditional techniques failed to find anything, no?

[degraded RAID]

Re: False positives

A tool is (mostly) only as good as its user.

[doublelayer]

Re: False positives

No, you can't. You can only assume that if you also know one of these:

1. You run them yourself and see that they don't find anything.

2. The people who wrote the code indicate that they always run all those tools and fix anything they find.

Many do not run those tools at all. Some who do don't bother to fix everything they warn about, either because they don't understand why it's a problem, don't have the skill or time to distinguish between true and false positives, or don't consider the bugs to be worth the time to proactively search for and fix. To some extent, that attitude can be justified because those tools, and for that matter LLMs, are capable of spitting out a ton of findings, some of which aren't real and some of which aren't important, and developer time is limited. However, a lot more of it is developers who don't know or care about security and don't even try to find problems until someone else reports one and announces that it will be made public so it would be great if they fixed it before then.

[Tom Womack]

I'm wondering how much of this is more connected to the NHS Biobank Data leaks, things like

https://www.bbc.co.uk/news/articles/clyedyn6pz7o

There is a difficult line to thread between 'your processing methodology ought to be open-source for good science' and 'there are great scientific benefits from processing large-scale NHS data'; checking your anonymised chunk of extracted data into your GitHub feels necessary by option 1, but when your analysis is on the metadata then there is almost always enough information in the metadata to deanonymise.

[bitwise]

Who made this decision

No word on who was in the meetings making this decision.

This will handicap development across government l.

As it is government departments all use libraries written to GDS standards and freely reuse fork and take over projects as needed.

This is going to be counter productive and result in slower and less secure apps

Who benefits from this?

I can imagine certain vendors being in favour of this got sure.

We paid for this stuff, it should not be closed.

[nijam]

Looking from a different viewpoint, whu did they choose github?

[Fara82Light]

Scope

Is it only the NHS that has taken this stance, or have any other UK Government agencies and ministries implemented a similar policy?

[Fara82Light]

Legal

The action might sound more plausible if the NHS were aware that it may or is about to be subject to a legal dispute over the ownership of all or part of the code it is making available as open source.