CIA used 'illegal, inaccurate code to target kill drones' | Hacker News Reader

CIA used 'illegal, inaccurate code to target kill drones'

theregister.co.uk

94 pointsmcantelon41 comments

> "They want to kill people with my software that doesn't work"

Let's see, we assassinate people without any trial, in a country we are not even at war with, and the obvious problem is that the software is a little buggy.

What is "buggy" here is the whole assassination system. When Bush wanted to dispatch special ops around the world to assassinate "terrorists", everyone was up and arms about it, When Obama does the same thing using drones, nobody seems to care, and the problem is a couple of bugs in the software.

There was a great piece whose identifying information I'm blanking on that explained how we stumbled into drone warfare. Basically, we tried doing snatch & grabs with commando teams for a while, and that kept generating one negative externality we weren't willing to tolerate. Illegal acts? No. Media pressure? No. Collateral damage? No. Diplomatic reprecussions? No. Prisoners.

Prisoners are a real handful. What are our options? Let them go? Well, no gain there. Put a bullet in the back of their head? Americans won't go for it. Try them in a civilian court? You get to expose sources & methods, they might walk anyhow (evidence tends to be testimony of equally unsavory characters plus something classified to the nth degree and not worth burning on Mook #473), and the successful outcome is they go plot their next terrorist act from a federal penitentiary. Try them in military court? All the fun of a civilian court, and the New York Times editorial board will make it sound like you took the bullet option anyhow.

You know what has never once happened as a result of a Predator drone strike? A live prisoner in US custody.

Interesting anecdote: up until WWII or so, American commanders had little problem dealing with enemy noncombatants on the battlefield. There was a brief trial followed by an execution.

Over time, the whole trial and such got to be counter-productive for the military, all that publicity and such. Warfare became much more a PR game. So commanders started emphasizing killing over capturing.

This trend has nothing to do with the war on terror. Eisenhower back in WWII sent out memos to his commanders saying he didn't want to see a lot of snipers being captured. The implications were pretty clear. Drone warfare is just the continuation of this larger trend.

American commanders had little problem dealing with enemy noncombatants on the battlefield.

Did you mean combatants? Typically noncombatants would refer to medical personnel, chaplains, etc. It'd be interesting to know if people like this were routinely tried and executed.

Perhaps he meant irregular combatants.

No, a chaplain in a military uniform is a combatant as far as the Geneva Convention is concerned, and must be treated as a POW.

The term more recently used was "unlawful combatant".

Though the Geneva Conventions do not state whether chaplains may bear arms, they specify (Protocol I, 8 June 1977, Art 43.2) that chaplains are noncombatants. In recent years both the UK and US have required chaplains, but not medical personnel, to be unarmed. Other nations, notably Norway, Denmark and Sweden, make it an issue of individual conscience. Captured chaplains are not considered Prisoners of War (Third Convention, 12 August 1949, Chapter IV Art 33) and must be returned to their home nation unless retained to minister to prisoners of war.

http://en.wikipedia.org/wiki/Chaplain#Military

I stand corrected :-)

But the point still stands - a chaplain in uniform on the battlefield wouldn't be tried and executed at a drumhead.

I suspect he means "spies and saboteurs" (to borrow the language of geneva), i.e. combatants who don't display a distinctive sign recognizable at a distance and openly carry weapons.

(The rationale: if one army can't tell the difference between civilians and soldiers, it puts civilians at greater risk. Thus, Geneva creates an incentive for not doing this.)

Good movie dealing with this in the Boer War: Breaker Morant (http://en.wikipedia.org/wiki/Breaker_Morant_%28film%29)

> You know what has never once happened as a result of a Predator drone strike? A live prisoner in US custody.

That makes a lot of sense. It is the optimal way of disposing of people on hidden black lists.

The other upside for them -- Americans don't put their lives in danger, unless we consider driving to Langley in DC traffic every day as "danger". So they never have to deal with that PR aspect of the operation.

That is entirely irrelevent.

Dropping a bomb from a plane or a UAV is basically the same thing in these wars. The enemies have no weapons capable of shooting down our fighters. Have we lost a single plane to enemy SAMs in either Iraq or Afghanistan? We've certainly lost some helicopters, but there really is no difference between an F/18 and a Predator.

If anything the UAV allows for more accuracy and a better view of the aftermath, the pilot of a manned plane is trying to get in and get out, the pilots of a UAV are watching before, during, and after the strike. The pilot of a manned fighter sees less of the actual attack and is even further removed.

The same can be said for ships at sea firing Tomahawks, and before that artillery. Once you get beyond the face to face fighting of infantry it just becomes more and more abstract until we are here today where the guys pulling the trigger are 3,000 miles away.

I don't disagree with what you said, but the real abstraction is the ability to shoot and not be shot at.

Even with "beyond the horizon" battles, two ships at sea are hitting each other. The difference between dying from a shell fired 20 miles away or one fired 1/2 mile away is nonexistent.

May I ask on what experience and evidence do you base all of your conclusions on? I assume you are or were a military pilot, or an analyst?

This whole conversation is way off-topic but the operations happen with the partnership of the Pakistani government (so the bit about "...in a country we're not at war with" is irrelevant) against military targets who are not fighting under any nation's flag.

The days of two armies lining up at opposite ends of a battle-field are gone.

> This whole conversation is way off-topic.

I think not when software is used to kill people. I refuse to sanitize and rationalize that fact away and happily discuss UAV realtime software development patterns.

The story was about UAV software, your agenda is extraneous. Otherwise any story can be used as a jumping off point for politics and righteous indignation.

Indeed. As someone who is:

1. Currently living neither in the US, nor the places it deploys these drones—and therefore unable to affect, or be affected by, the US's foreign policy re: drones;

2. and who rationally values their time;

I can't see any benefit to investing said time in analyzing US drone deployment policy. The technology of drones, however, is fascinating.

There's always the possibility that (1) will one day change. For example, when, in some state of emergency, the US decides to back one political strength over another. I'm not military or historical expert, so I might be wrong, but I could easily see this happening in countries like Mexico.

Good point. The only problem is the Netezza appliance is definitely not used by drones, for controlling drones, or anything directly related to their operations. The Register headline (and the many copies) are typically hyperbolic.

The software is used to determine target locations,which are then fed to the drones for execution. I'd say it's pretty damm relevant when someone decides to hack something up real quick and says they don't mind using it knowing that it's giving erroneous data with several meter deviations. I bet you wouldn't find it hyperbolic if you were, let's say, 10 meters away from someone being located with that software.

Yes, it could be better, but in the grand scheme of things it's a major improvement over what came before. Less than a century ago, in WWII, bombers often missed their target by miles. Miles. We are orders of magnitude more accurate, leading to smaller amounts of explosive used, and fewer people being killed by accident.

No, it's never good when an unintended person is killed, but let's keep things in perspective.

I think not when software is used to kill people.

This is not new. Software is also in most components of military weapons and vehicles: in missiles, tanks, submarines, torpedoes, artillery, helicopters and fighter jets, etc. What's different about UAVs?

It's important to remember that neither Bush nor Obama are king of the U.S. government. Our newspapers, universities, and civil servants turned on Bush because of his entry to Iraq, acting out of step with the rest of the government (which includes a lot more than people think). Anything they said to reduce Bush's power was fair game. You, like me, probably fell for it. People in the government generally don't have a problem with the government's power and actions. They did pull off a great trick though, assigning everything bad to Bush for six years.

> It's important to remember that neither Bush nor Obama are king of the U.S. government.

That is true. I was just using the names as shorthand for the governments they run.

> You, like me, probably fell for it.

Unfortunately that is true. Since then I started to pay more attention to how propaganda works in US.

> They did pull off a great trick though, assigning everything bad to Bush for six years.

Yes that was brilliant of them. It looks like it worked rather well.

Could you elaborate on this? Normally I don't care about politics, but this does sound interesting.

I'm pretty sure I stole that interpretation from this guy: http://moldbuggery.blogspot.com/ - I spent a while looking for the exact article, but didn't find it. Most of them are very interesting anyway. He's a fascinating character, and even if you don't like him, he refers to great primary sources. But do watch out for three of his big biases: 1) assuming everyone is as smart as he is; 2) growing up as a "Brahmin" (his neologism); 3) preferring that the U.S. solve its own problems somehow (patriotic but very difficult).

Thanks a lot, this looks very interesting. Maybe on the overkill side, but I'm kinda hooked.

Let's not forget that Clinton's favourite tactic was a cruise missile too. Remember Sudan?

This is the problem with success, it has many different meanings and some of then are evil. When using technology to kill people it should be carefully studied what is the meaning of success.

You really can't tell a lot from this article. It all comes down to what this piece of software does. A faulty component in the UAV doesn't necessarily mean that the weapon lands off target.

They definitely aren't modifying the guidance system on the weapon - those weapons are not built by the CIA. It sounds like this piece of software helps them correlate coordinates for targeting from one computer system to another. That may be bad, it may not.

(Also, in my experience the vetting of designs is pretty rigorous, but I work on the mechanical side and our customer is DOD not CIA so your results may vary.)

> A faulty component in the UAV doesn't necessarily mean that the weapon lands off target.

Someone on reddit pointed out that even if it does, it doesn't matter. If the kill radius is 20 yards, and the bugs mean that it has an error of up to 5 yards... that's not really significant.

edit: Unless, of course, you happen to be bombing something next to an elementary school, a mosque, or a hospital.

I would argue that any time accuracy (or any feature) is expected, then a lack of said feature does matter. You never know when someone somewhere could be relying on the fact that they can get within 5 yards.

Look up "circular error probable" (CEP). When they say "accurate to x meters", they might well mean that 50% of munitions land in an x meter circle --- the other 50%? anywhere (maybe gaussian distro). So all this talk of x meter accuracy is quite misleading.

That may be true, but (assuming a normal distribution [1]) that would also imply that 99% of shots land within 2.6x meters (assuming I didn't mess up my arithmetic).

Whoever is doing the aiming gets a big red circle that encloses the margin of error. If you are off by 5 meters, that big red circle is in the wrong place. The pilot expects a 1% chance of error, in reality it might be 4% (i.e., 4x as many screwups).

[1] Of course it isn't a normal distribution, but this only changes the programming details.

This article is missing some key facts.

Netezza appliance has nothing to do with controlling drones. The tenuous link in the article is established by a failure to understand the intel use of the word targeting (gathering info on a target), as opposed to the aiming a missile military sense. Why would a drone use a data warehouse plugin to control anything operational? It wouldn't. Could a coordinate come out of the system- sure, but just a coordinate that someone else put in there.

Netezza sued IISi in 2009 for failing to deliver the port of their software. It looks like they didn't seem to get much traction with that, made some surprisingly clumsy document releases, showing where they laid down some stupid spy BS to try and push IISi to deliver or give up, or at least hand over the code so the customer could get it working. http://www.thestreet.com/print/story/10810646.html

Now, we're at the third stage, where the countersuit is claiming reverse engineering (considering that Netezza probably wrote the specs, hard to prove) and Netezza is getting bought by IBM, perhaps to get their legal team aimed at the issue...meanwhile, it gives everyone a good excuse to talk about drones.

I love government contracting. First we pick a schedule, then we pick a technology, then we decide who gets the contract and who will supervise the execution of said contract. The best part? When something gets delivered, no matter how sad a state of affairs it is code-wise, no matter how difficult or expensive it will be to bring up to any sensible standards of maintainability, they won't pay for it. No, they'll only pay for a new feature.

This is why government contracting is fun. You never know when a story like this is going to come out and if you're going to be involved with it.

Let me get this straight, the drone is inaccurate by the worst case 13m. So the drone itself is off by at most 13m, while being 26,000ft in the air. This hardly seems like a major issue to me, according to wikipedia the Predator drone is equipped with hellfire munition, which have their own laser guided trajectory. Depending on the type the hellfire munition has approximately 8000ft range, so something tells me the drone itself doesn't guide the hellfire to the ground.

But 13m? I'm no munitions/combat expert, but I assume 13m discrepancy, even if it does apply to the hellfire itself, won't make much of a difference. Something is going to get blown up anyway, even if it is off by 13meters the target will essentially no longer be there.

The CIA does not seem to be having very good press karma with their drones, if I recall there was a few issues with data stream encryption, and software that allowed hostiles to intercept and view UAV downlinks in the battle field. Perhaps its just tempting to write a story about such a high profile target, but it seems partly like some corner cutting is going on on the CIA's end.

I don't really care if it's illegal. But inaccurate? That's not good.

That's the main problem with American foreign policy. Win-Lose proposition is not scalable in globalized internet world.

Reply on news.ycombinator.com