Parcel level bot detection

[xDancingStarx]

1 hour ago, Phil Deakins said:

LL gave us a means of identifying scripted agents, and some people want to make use of it. It has nothing to do with anyone else.

It's important to be correct here, and your sentence should be "LL gave us means of detecting identified scripted agents". As it was written in the initial post. LL did not give us means of identifying scripted agents. I'm just clarifying this to prevent people from reading this thread only partially and being misinformed.

Edited August 19, 2023 by xDancingStarx

[Qie Niangao]

5 hours ago, xDancingStarx said:

The whole severity of the "problem" of intrusive bots can be summed up by the fact that it needs scripts logging them in order to grasp their numbers. If there weren't region enter/leave messages, we wouldn't even be talking about this. It's NOT the case that you stand in your parcel and there's a bot appearing next to your avatar every 10 minutes.  […]

Thing is, these logging scripts are showing it's vastly closer to that than any of us could ever have imagined. (Unless it's not. More below.)

I actually agree that in most cases it's best not to ban these things, lest they instead simply go undercover as undeclared scripted agents. If they started sticking around longer than a few seconds, then yeah, begone with 'em, but otherwise I'm more interested in counting and exploring their identities than trying to remove them.

Also, all my parcels are adjacent protected Linden land, so the most I could do would be to move them all over there (where most of them are already). They'd be about the same annoyance rezzing a few meters from wherever they rez now, so in my case I'd only inconvenience the laziest of botrunners—and then only the ones playing by the rules.

The thing is, so few of them confess why they're rezzing anywhere at all. Of course they're not required to reveal anything, but if they don't, who wouldn't start from the assumption they have something to hide? 

Now, though: How bad is this AGENT_AUTOMATED false positive rate? I mean, the ones I'm seeing are definitely bots: they flit in and out too quickly to be anything else. Am I just "lucky" and misidentified ones don't show up? (Actually, that's pretty possible; other than road, rail, and water travelers, there's not much traffic through the regions where I have land.)

[xDancingStarx]

1 hour ago, Qie Niangao said:

I actually agree that in most cases it's best not to ban these things

If a user annoys me, I would ban them. That's completely legit. The problem here is that some people already feel annoyed by any user (bot or no bot) who enters their parcel even when they are not present. They look at their visitor log, and they get annoyed. I would have no problem agreeing that there's a bot problem if it actually existed and people weren't just looking at their logs or region messages in order to get annoyed. Furthermore when security devices get the simple option to automatically kick avatars declared as scripted agents, this would be something that sounds like making sense for any user to activate. Kick bots? Hell yeah. We're already on a path that I had predicted months ago, that this functionality to see who's a scripted agent is becoming completely useless, since people are gonna get encouraged more and more to not declare their avatars as scripted agents. It's probably unavoidable at this point of time. And I'm not defending bots that enter private parcels by any means. But I'm also not considering it a massive problem when anyone enters a private parcel when nobody is there. Bots will not disappear. Never, and this is just a fact. And right now people are starting to do everything to make sure that nobody knows anymore who a bot is.

Edited August 19, 2023 by xDancingStarx

[Gabriele Graves]

2 hours ago, Qie Niangao said:

Now, though: How bad is this AGENT_AUTOMATED false positive rate? I mean, the ones I'm seeing are definitely bots: they flit in and out too quickly to be anything else. Am I just "lucky" and misidentified ones don't show up? (Actually, that's pretty possible; other than road, rail, and water travelers, there's not much traffic through the regions where I have land.)

It would be helpful to know how many of these had been logged in at some point in the past with SpeedLight and/or bot software that might have corrupted the results.

There is definitely something for LL to investigate here or it all becomes pointless.

All of the "scripted agents" I've logged on my land have been in and out in a few seconds which tells me that I'm probably not seeing any false positives.

EDIT: I've been logging for about a month and have seen 210 scripted agents so far according to my logs.

Edited August 20, 2023 by Gabriele Graves

[Phil Deakins]

7 hours ago, xDancingStarx said:

I would have no problem agreeing that there's a bot problem if it actually existed

It doesn't exist. Not for you. But, as we have seen in this and other threads, it does exist for some people. Whether or not their annoyance is justified is nobody else's business. It's their world, not yours.

Edited August 20, 2023 by Phil Deakins

[Phil Deakins]

7 hours ago, Gabriele Graves said:

It would be helpful to know how many of these had been logged in at some point in the past with SpeedLight and/or bot software that might have corrupted the results.

Love asked that question in another thread. Of the 4 of 16 that were wrongly identified as scripted agents, none had been logged in with Speedlight.

On the other hand, some of us who were not wrongly identified as scripted agents, had been logged in with Speedlight. One was logged in with Speedlight at the time I did the 16 avs test, and it wasn't wrongly identified as a scripted agent.

So I've come to the conclusion that it's nothing to do with Speedlight.

Apart from me, they had all been logged in with bot software in the past. Most of them were logged in with bot software for the test, but only 4 of them were wrongly identified. So bot software isn't the reason either. At least not the bot software that they've used in the past. I've only ever used my own bot software for any of them.

Even if it is something to do with Speedlight, bot software, or something else, bit 15 wasn't available for a scripted agent flag, so it's bugged.

Edited August 20, 2023 by Phil Deakins

[Phil Deakins]

This is interesting. I was contacted by someone who has been running scripted agent detectors for some time. The one test that I hadn't done was test the false positive avs on other land. I couldn't imagine how the land could make a difference, and it doesn't. I took my false positive avs to his detector and it also wrongly reported them as scripted agents. Then I wondered if it's something to do with just my accounts, but he has also detected false positives. Nothing like the percentage I have, but false positives just the same.

They are out there........lurking

[Qie Niangao]

So have you and Peter exchanged lists of false positives? Does any theory remain as to what specifically has this bit flipped on the false positive accounts?

With the false-positives so hard to find, it seems too soon to run a probably destructive test, but it might be interesting to know if the bit clears if one of the false positive accounts declares themselves a scripted agent and then removes that declaration. (Un-declaring does clear the bit normally, perhaps after a relog, as my Bellisseria-deprived alt can attest.)

Just a total shot in the dark: I notice the "Wicked Leigh" account dates from 2007 and it was 2009 when they added AGENT_ALWAYS_RUN, the most recent addition to llGetAgentInfo() before AGENT_AUTOMATED. Remember that Aristotle/Integrity "identity verification" hoax they tried for a while when the Adult Content policy was new? Might they ever have used (or tried using) a flag in the AgentInfo bitvector for those accounts who went through that? or maybe accounts that got in some identity limbo by those scammers? If it's anything like that, the false positive accounts would all be from a certain date or earlier.

[Love Zhaoying]

14 minutes ago, Qie Niangao said:

So have you and Peter exchanged lists of false positives? Does any theory remain as to what specifically has this bit flipped on the false positive accounts?

With the false-positives so hard to find, it seems too soon to run a probably destructive test, but it might be interesting to know if the bit clears if one of the false positive accounts declares themselves a scripted agent and then removes that declaration. (Un-declaring does clear the bit normally, perhaps after a relog, as my Bellisseria-deprived alt can attest.)

Just a total shot in the dark: I notice the "Wicked Leigh" account dates from 2007 and it was 2009 when they added AGENT_ALWAYS_RUN, the most recent addition to llGetAgentInfo() before AGENT_AUTOMATED. Remember that Aristotle/Integrity "identity verification" hoax they tried for a while when the Adult Content policy was new? Might they ever have used (or tried using) a flag in the AgentInfo bitvector for those accounts who went through that? or maybe accounts that got in some identity limbo by those scammers? If it's anything like that, the false positive accounts would all be from a certain date or earlier.

I like your idea!! If the bit was EVER used for anything "not documented" but not cleared..(like if ANY third-party viewer had set it, or if it had been used by an undocumented feature long ago)..!

Edited August 20, 2023 by Love Zhaoying

[Phil Deakins]

1 hour ago, Qie Niangao said:

So have you and Peter exchanged lists of false positives? Does any theory remain as to what specifically has this bit flipped on the false positive accounts?

Not both ways, but Peter knows mine now. I have around 70 accounts and I've only tested around 25 of them. I don't see any need to test the rest. I could test them but it would a daunting task as most of them would log in to various parts of SL and I would have to bring them to my place. I think that what I have is enough
 

1 hour ago, Qie Niangao said:

With the false-positives so hard to find, it seems too soon to run a probably destructive test, but it might be interesting to know if the bit clears if one of the false positive accounts declares themselves a scripted agent and then removes that declaration. (Un-declaring does clear the bit normally, perhaps after a relog, as my Bellisseria-deprived alt can attest.)

I did set one of the false positives to being a scripted agent, and then back to being a human, but it made no difference. That's a bit odd really, as setting and resetting the flag ought to have left that account's bit reset. The account has been logged in multiple times since then, and it's still returned as a scripted agent; i.e. that bit is set. I can see it.
 

1 hour ago, Qie Niangao said:

Just a total shot in the dark: I notice the "Wicked Leigh" account dates from 2007 and it was 2009 when they added AGENT_ALWAYS_RUN, the most recent addition to llGetAgentInfo() before AGENT_AUTOMATED. 

Two of the false positive accounts were created in 2007, and 2 were created in 2010. Wicked Leigh was a paid-for alt. I have many other accounts that were also created in 2007/8 that were actually used as bots. Not the 2010 ones though. I have plenty of those too but they were created for their Picks. In other words, there is no similarity between them, except that they are old. I haven't tested all my accounts but, of the ones I've tested, other accounts  similar to those aren't returned as scripted agents. Maybe I should test all my accounts and check their creation years.
 

1 hour ago, Qie Niangao said:

Remember that Aristotle/Integrity "identity verification" hoax they tried for a while when the Adult Content policy was new? Might they ever have used (or tried using) a flag in the AgentInfo bitvector for those accounts who went through that? or maybe accounts that got in some identity limbo by those scammers? If it's anything like that, the false positive accounts would all be from a certain date or earlier.

I've never heard of that.

Whatever the reason, bit 15, the one that's used for AGENT_AUTOMATED, was or is being used for something else, and since I tested setting and unsetting the scripted agent status of Wicked Leigh, without curing it, it seems like it's currently being set for some other reason, which itself could be an old reason that's been forgotten about.

ETA:
I logged more in and found another one - created in 2007. White I was doing it I found many that were created in 2007, many in 2010, and a few in between, that weren't wrongly identified as scripted agents. I also found a couple that were correctly identified as scripted agent.

Edited August 20, 2023 by Phil Deakins

[Qie Niangao]

2 hours ago, Phil Deakins said:

I've never heard of that. [the Aristotle/Integrity identity verification thing the Lab tried to use for adult age verification]

A couple links might jog your memory:

• A Massively article from 2007 listing some of the more hilarious reasons the practice was widely derided.
• A forums thread from 2011 where it was still a subject of discussion, so presumably still in use at that time to determine who could and couldn't enter "Adult"-rated parcels. (Now they just use the date of birth supplied at account creation, which is supposedly standard practice in the games industry.)

Just to be clear, my speculation about an  llGetAgentInfo connection would not have everybody from that era get a tainted AGENT_AUTOMATED bit, just those who tried but failed to verify (common for non-US users, but not exclusively so), or some other age-verification status they tried to record.

It all might seem unlikely, given that Wicked Leigh's tainted soul could not be saved. On the other hand, that's weird anyway: there's evidently some process that actively populates the AgentInfo bitvector from bogus information, "being set for some other reason", even after it should be cleared.

[Love Zhaoying]

I think if only "older" ("not brand new") accounts give false positives it will help support certain theories.

[Silent Mistwalker]

6 hours ago, Phil Deakins said:

This is interesting. I was contacted by someone who has been running scripted agent detectors for some time. The one test that I hadn't done was test the false positive avs on other land. I couldn't imagine how the land could make a difference, and it doesn't. I took my false positive avs to his detector and it also wrongly reported them as scripted agents. Then I wondered if it's something to do with just my accounts, but he has also detected false positives. Nothing like the percentage I have, but false positives just the same.

They are out there........lurking

Sounds like whatever code LL uses to remove the scripted agent flag is failing at times on accounts that are no longer used as bots. That is if LL has bothered with any kind of code like that. 😬

[Phil Deakins]

34 minutes ago, Qie Niangao said:

A couple links might jog your memory:

It did jog my memory, thank you. In fact I went through that process myself. Just this account though, none of the others.

[Phil Deakins]

38 minutes ago, Love Zhaoying said:

I think if only "older" ("not brand new") accounts give false positives it will help support certain theories.

That the bit was allocated for some reason in the dim and distant past, and then forgotten about? That's what it's looking like to me. And it's still effective because registering and deregistering doesn't fix it. If I register a false positive account as a scripted agent, and then deregister it so that the account says it's controlled by a human, the flag is still set (scripted agent) when AGENT_AUTOMATED is returned.

But thinking about that, where does the account page get its information from when it says that the account is controlled by a human? Is it the same database that llGetAgentInfo() gets it from? If it is, they can't show different things OR something is actually setting that bit in the process of sending the agent info.

I only have 2 accounts that are new enough to have a Resident last name, and they come back ok. But it's such a minuscule sample that we can't read anything into it about the age of the accounts.

We're not going to get to the bottom of it, and it's up to LL to figure it out - if they  actually want to. They should because it's an honest-to-goodness bug.

Edited August 20, 2023 by Phil Deakins

[Phil Deakins]

58 minutes ago, Silent Mistwalker said:

Sounds like whatever code LL uses to remove the scripted agent flag is failing at times on accounts that are no longer used as bots. That is if LL has bothered with any kind of code like that. 😬

I don't think that bots or no bots is behind it. If LL has a system where known or suspected bots were flagged somewhere, and it's being looked at when returning agent info, so that matches set the AGENT_AUTOMATED flag when it's requested, a lot more of my accounts would be flagged like that. When I used bots, they were used in much larger bunches than a mere 4 or 5, and they would have been flagged.

But I am leaning towards the idea that something from the past is checked when AGENT_AUTOMATED is requested, and, if matched, bit 15 is set for the return but not actually in the account's data. The account page says 'human' but the returned agent data says 'scripted agent', so it's looking like the code checks something else, and modifies bit 15 accordingly when returning the requested AGENT_AUTOMATED data.

Edited August 20, 2023 by Phil Deakins

[Paul Hexem]

I saw some bots sitting together in a Linden home the other day when I had my boat out. Presumably their owner is premium, owns the home, and has them in the home group.

It's a great real world example of that use case I pointed out right from the start, I knew people would run into it.

And now that we're seeing more detectors cropping up especially reporting false positives, I think we're going to start seeing more and more bots not registered as scripted agents. When it's less annoying to ignore the rules than follow them, what do you think people will do?

Edited August 20, 2023 by Paul Hexem

[Phil Deakins]

I don't really see a problem. Since the 'scripted agent' status came in a long time ago, there have been many posts about us not being able to recognise them. The posts were about bunches of avatars that appeared to be traffic bots but there was no way of knowing for sure. Of course, traffic bots wouldn't be registered as scripted agents, or they wouldn't be traffic bots . So what difference does it make if a bot is registered as such or not?

There is no requirement for bots to be registered as scripted agents. It's not against the rules for them not to be. I think that LL introduced the AGENT_AUTOMATED option so that people could report bunches of bots that appear to be traffic bots but aren't registered as scripted agents, instead of reporting all groups of avatars that just may be traffic bots. It's a time-saver for LL

[Silent Mistwalker]

45 minutes ago, Phil Deakins said:

I don't think that bots or no bots is behind it. If LL has a system where known or suspected bots were flagged somewhere, and it's being looked at when returning agent info, so that matches set the AGENT_AUTOMATED flag when it's requested, a lot more of my accounts would be flagged like that. When I used bots, they were used in much larger bunches than a mere 4 or 5, and they would have been flagged.

But I am leaning towards the idea that something from the past is checked when AGENT_AUTOMATED is requested, and, if matched, bit 15 is set for the return but not actually in the account's data. The account page says 'human' but the returned agent data says 'scripted agent', so it's looking like the code checks something else, and modifies bit 15 accordingly when returning the requested AGENT_AUTOMATED data.

That's what I'm trying to say but don't have the language to know the terms. The correct data isn't being transmitted when an account is removed from being a scripted agent. So, the data is either blocked somehow or isn't being transmitted because it's "in the wrong place" or nonexistent.

In any case, I think you guys are on the right track and it's definitely a bug.

Edited August 20, 2023 by Silent Mistwalker

[Phil Deakins]

A thought is meandering through my mind 

I did use a lot of bots at one time, and I'm wondering if the false positives accounts were reported and flagged back then as individuals. I registered some obvious ones but not others, and odd ones might have been reported. A Linden might have simply flagged the reported one, and not looked for others. It's just a thought.

@Silent Mistwalker I'm back to not altogether ruling out the bots possibility 

Edited August 20, 2023 by Phil Deakins

[Silent Mistwalker]

6 minutes ago, Phil Deakins said:

A thought is meandering through my mind 

I did use a lot of bots at one time, and I'm wondering if the false positives accounts were reported and flagged back then as individuals. I registered some obvious ones but not others, and odd ones might have been reported. A Linden might have simply flagged the reported one, and not looked for others. It's just a thought.

@Silent Mistwalker I'm back to not altogether ruling out the bots possibility 

Sorry about that. It's just how my messed up brain works. 😄

[Phil Deakins]

Sorry about what? The laugh? I thought it fit beautifully

I've submitted a support ticked for Wicked Leigh, hoping to find out why it happens. There was no remotely suitable option in the Account Issue list and I had to pick one that's unsuitable. I hope it goes through.

[Silent Mistwalker]

2 minutes ago, Phil Deakins said:

Sorry about what? The laugh? I thought it fit beautifully

I've submitted a support ticked for Wicked Leigh, hoping to find out why it happens. There was no remotely suitable option in the Account Issue list and I had to pick one that's unsuitable. I hope it goes through.

No. Sorry about the way my brain works. 🤭

Good luck! Let us know how it goes please.

[Phil Deakins]

I certainly will. I haven't written all these posts only to keep quiet if I find anything  out 

[Love Zhaoying]

52 minutes ago, Phil Deakins said:

A thought is meandering through my mind 

I did use a lot of bots at one time, and I'm wondering if the false positives accounts were reported and flagged back then as individuals. I registered some obvious ones but not others, and odd ones might have been reported. A Linden might have simply flagged the reported one, and not looked for others. It's just a thought.

That would be a teensy bit ironic!

I mean, if you were seeing a lot of false positives on your own avatars, because they "had been bots" at one time.

I am waiting with *bated breath on the results of this one!

Edited August 20, 2023 by Love Zhaoying
*bated, not waited