[Here is an article about an on-going court martial in the UK. The sort of situation and allegations discussed are well-known to RISKs readers, so I have provided the quote essentially just for the record. Brian Randell Computing Laboratory, The University, Newcastle upon Tyne, NE1 7RU, UK EMAIL = Brian.Randell@newcastle.ac.uk PHONE = +44 91 222 7923 ] COMPUTER BLAMED FOR SEA COLLISION (The Independent, 5 Jun 1992) A Trainee submarine commander yesterday blamed a computer error for an accident which sank a trawler and killed four Scottish fishermen. Lieutenant Commander Peter McDonnell told a court martial at HMS Drake in Plymouth that he trusted HMS Trenchant's computer system when it told him he was at least three miles away from a possible collision with the Scottish trawler Antares. He said his generation of submariners preferred to rely on the computer rather than a manual plotting system which a senior submarine captain earlier told the hearing was a more trustworthy method in busy waters. Four men died in November 1990 when the Antares was dragged to the bottom of the Firth of Clyde by HMS Trenchant. Lt Cdr McDonnell, 33, from Glossop, Derbyshire, had just completed the last exercise of a six-month command course known as the Perisher when the accident occurred at 2.18am. He denies six charges of negligence. Yesterday he told the hearing that he had not even known that Trenchant had passed close to the Antares and another fishing boat five minutes before he ordered the submarine to turn around and head back towards them. The hearing continues today. [Ross.Anderson@cl.cam.ac.uk found most of that in The London Times as well.]
SORRY, BUDDY - IT SAYS RIGHT HERE THAT YOU'RE DEAD (Indianapolis Star, June 3, 1992) And you think you've had trouble dealing with apathetic bureaucrats? Meet Eugene Smith of Doylestown, PA. The healthy 33-year-old has spent the past 2.5 years convincing authorities he's not dead. The frustrating error cost him his driver's license and his job. He still can't get a license, and he's still fighting nine traffic violations that he says aren't his. Smith traces the trouble to the theft of his wallet in 1988. He believes the thief used his driver's license, racked up violations that led to the license suspension, then died in a traffic accident. In February 1990, a police officer stopped Smith and told him his car regis- tration was expired and that state computer records showed he was dead. "He said I was dead, and because of that I was not allowed to drive," said Smith. "I agreed that it would be [a] hazard for a dead person to be driving." Life isn't easy for an officially dead man. Without a license, Smith lost his job as a driver for a warehouse. Without that job, he had to find a cheaper place to live and take a job nearby, at a deli. Being an officially dead taxpayer, no one in the state capital took him seriously. "I would call and I could hear them say, `Oh, this is that guy again,' and I could hear them laugh and they would say nobody there could help me," Smith said. Finally, Susan Rakus, an aide to Democratic U.S. Rep. Peter Kostmayer, took his case and persuaded the state motor vehicle agency to resurrect Smith [isn't this against separation of church and state?]. But Smith still can't get a license — he's still accused of a string of years-old traffic violations. "Obviously we dropped the ball on this," Rick Schoen, state transportation department spokesman, said Tuesday. William Joseph Winn firstname.lastname@example.org
There has been a discussion going on in the Telecom forum about *67 which TOGGLES(!!!!!) the caller-id blocking state of a phone line — at least in those areas with caller-id blocking. The rationale for requiring caller-id blocking in some states is that there are situations where disclosing one's location might be life-threatening as in the case of a shelter for battered women or maybe a protected witness. Of course, there are also normal privacy considerations. If one always was sure of the default state of the line one was using a toggle might work. But there is no way to determine the state beyond faith that the telco's computer is exactly synchronized with one's expectations and that one has is using the assumed CO lines on multi-line systems. If one is a visitor, all bets are off. As from plain errors made in the business office or at the CO, one reader pointed out that one some switches reloading the software loses the settings. Another reader pointed out that *67 isn't an accident but the specified behavior. The stupidity (the word risk doesn't do justice to the situation) is obvious. I'm more puzzled about how it came about. I generally lean towards incompetence as an explanation rather than conspiracy but since some of the rationale for requiring caller-id comes from public safety considerations, I'm surprised that no one has challenged this approach as failing to satisfy this requirement and, by providing the illusion of caller-id blocking, might increase the risk. While on this subject, there is also the issue of access control over information passed via signalling protocols. Telcos are assumed to have full access and subscribers none. But some organizations can act as their own telcos. The MIT ISDN switch comes to mind. Which side of the protection barrier are they on? ANI is similar to caller-id but is nonblocked and delivered when calling an 800 #. This means that if I give out my personal 800#, I will eventually (on the next bill) get their #.
In today's Wall Street Journal, there was a feature piece on a slot machine tournament in Atlantic City. The problem was that the machines were returning a 70.6% payoff rather than the 96.4% planned. "After the tournament ended and the prizes were awarded, the manufacturer called back to report that the two kinds of chips it shipped were incompatible with each other". Aside from all the issues of how this might have happened, the real danger is soft failure that are hard to detect. The only reason someone even looked for a problem was the unique circumstances of a tournament which provided an environment to notice the statistical anomalies Apparently there is no constant checking to see that the statistical results match the predicted results. The *67 (above) and this story both illustrate a risk of not understanding the philosophical (as well as engineering) concept of closed-loop systems, i.e., those with feedback so that one can determine the result of an action. This is a lesson that should feedback to nontechnology systems also. [Chuck Weinstock <weinstoc@SEI.CMU.EDU> also noted the slot machine saga, as did Roland Ouellette, who added the note that follows. PGN
This makes me wonder if anyone actually tests these machines: people at the factory or regulators at the casinos. Also would this sort of error be noticed only with an event like this and ordinarily go undetected? Roland Ouellette
I've had up on my door an article from the 4/14/92 Weekly World News an American tabloid) with a headline: "Girl, 13, kidnaped by her computer!" Here is an excerpt: A desperate plea for help on a computer screen and a girl vanishing into thin air has everyone baffled --- and a high-tech computer game is the prime suspect. Game creator and computer expert Christian Lambert believes a glitch in his game Mindbender might have caused a computer to swallow 13-year-old Patrice Toussaint into her computer. "Mindbender is only supposed to have eight levels," Lambert said. "But this one version somehow has an extra level. A level that is not supposed to be there! The only thing I can figure out now is that she's playing the ninth level --- inside the machine!".... Lambert speculates that if she is in the computer, the only way out for her is if she wins the game. But it's difficult to know for sure how long it will take, Lambert said. "As long as her parents don't turn off the machine Patrice will be safe," he said. "The rest is up to her." Why am I posting this to comp.risks? Do I really think there is a risk of people being kidnaped by computers? No (although at times, when I work on my thesis, I wonder.) The risk is the misinformation people receive about computers. I don't worry too much about the WWN, but I was concerned about an educational show I watched last night, Mathnet, based on a segment of the PBS educational television show, Square One. Mathnet is a spoof of the detective show Dragnet, and the detectives use math to solve crimes. So far, so good, but on last night's episode, the crime they solved was the kidnaping of a baseball player whose disappearance had been unnoticed because he had been replaced by an android which had been able to talk and play baseball. An educational show would not show space aliens or magic, so the implication of including human-like robots is that they are technically feasible. Similarly, when I recently visited Epcot, an amusement park that is supposed to be educational, the computer exhibit featured an electronic character that was able to understand and even physically transport its human companion. I expect (and enjoy) such unrealism in tabloids and in science fiction, but it should not appear in educational settings. I suspect that a large percentage of people, if asked, would say that a robot could currently be built that could pass as human, based on all the misinformation they receive. Ellen Spertus
Isn't it just like our technocratic society to react to such an accident, caused by a completely unnecessary luxury becoming too complex, by making it even more complex? Wouldn't the simpler solution be to ban automatic windows or even power windows instead of requiring another safety interlock? Nobody needs such things but, unfortunately, there are car models in which you can't get an ABS (good thing) without buying power windows (artificially induced desire). I told my dealer that I was willing to pay extra for manual windows, but could not get them.
Viruses are a dime a dozen nowadays, but I thought this one was of particular interest (though I do have to wonder if the issue of "Computing" magazine was from April 1st...). >From Communications of the ACM, June 1992 (vol.35, no.6), page 10: Barry's Bug... Viruses, as we all know, can play strange and frightening games with computer-based data. Now, "Computing" magazine has reported a new strain that plays some strange, and yes, frightening music. It's called the Barry Manilow Virus - a phantom bug that's infiltrating a growing number of computer systems, scaring users with such tunes as "Mandy" and "Copacabana." The virus is a collection from the singer's "Greatest Hits" album. Once detonated, the virus spins out a continuous stream of Manilow's million sellers. Experts are working feverishly on an antidote for this plague. — Eric Haines
My family is enrolled in AT&T's World Reach-out plan. This plan provides discounted calls to many countries throughout the world during designated times, including what used to be West Germany. However there are no discounts to what used to be East Germany (GDR). At our house, we call Germany (the western part) a lot. Yesterday we received our May phone bill from Ohio Bell. I noticed that after around May 5 our calls to Germany did not have the Reach-out discount. Also, the designation of the location called was changed from "Ger Fed Rep" to "Germany". I called AT&T, and a rate adjuster told me that the problem was with Ohio Bell's billing software. It seems that their software was keying off the "Ger Fed Rep" to apply the Reach-out discount, not the country code (49). Thus, in May, when AT&T decided to change the designation "Ger Fed Rep" to "Germany", the software broke. AT&T credited me the difference, which was $21.00. I wonder how many other phone companies will have the same problem and how many other people will be affected. Adnan Yaqub (email@example.com) Allen-Bradley Company, Inc., 747 Alpha Drive, Highland Hts., OH 44143, USA Phone: +1 216 646 4670 FAX: +1 216 646 4484
With the increasing amount of casual communication these computer networks (like usenet) are encouraging, this namespace collision situation is likely to increase. I recently experienced this. A few months ago, I posted an article to comp.realtime which quoted the US GAO report on the Patriot missile failure. Somebody read it there, and reposted it to the widely-read comp.risks forum. Shortly thereafter, I received an e-mail message from another person named Fred Roeber. He works for Raytheon, the makers of the Patriot system! His father, also named Fred Roeber, also works for Raytheon. He saw my article, and immediately fired off letters to his superiors, alerting them that the posting was *not* inside information from either one of them, but public information from someone with the same name. Luckily, it seems that no harm has come from this. In fact, two branches of a family that hadn't known about each other can now fill in some gaps in the family tree. But if one of his superiors had seen the article first, and acted prematurely; or if the GAO or I had made a mistake that Raytheon might have considered slanderous, the results could have been much worse for him. The RISK seems to me to be that if we do not realize just how large this increasingly popular global community is, we may mis-estimate the probability of such a collision, and make mistaken assumptions about identity. Frederick G. M. Roeber | CERN — European Center for Nuclear Research e-mail: firstname.lastname@example.org or email@example.com | work: +41 22 767 31 80 r-mail: CERN/PPE, 1211 Geneva 23, Switzerland | home: +33 50 42 19 44
A little over a year ago, I was hounded by a collection agency for debts owed by Richard Simkin, a car dealer in northern Illinois. It took about a month (and a letter to the Better Business Bureau) to convince the agency that I wasn't their man. Late last fall, I applied for and received a Discover Card. About 4 months later, Discover Merchant Services decided that my name matched that of Richard Simkin of Roselle Motors and tried to collect his debts from me. The pattern was to leave a phone message, or send a letter, telling me to call Ranee. Phone messages (especially the first time, when all this was news to me) never said why I should call. When I would call, Ranee was never in the office, so I'd end up talking to someone else. I'd explain that I wasn't a car dealer, and that they'd mixed me up with somebody else. They'd promise to take care of the problem; once a supervisor told me that I shouldn't have gotten a letter at all--he couldn't even figure out how it got to me, since my address wasn't on the record of the delinquent merchant--and I should ignore it. I've cancelled my account now, hoping that if there's no customer record, they won't match it to their merchant record. I'm told that Discover policy requires more than a matching name to claim that two records represent the same person; and that by that policy, my record does not match that of the car dealer's. Computer Risks: - Computer programs don't always reflect company policy. - Flexible tools (such as a database query language and mail merge) provide an easy means to act on wrong assumptions, and don't always leave audit trails the way tailored applications can. Rick Simkin UUCP: uunet!dlogics!rsimkin Datalogics, Inc. INTERNET: firstname.lastname@example.org 441 W. Huron St. PHONE: +1 312 2664437 Chicago, Illinois 60610-3498 USA FAX: +1 312 2664473
>It all boils down to what your definition of "virus". My definition of "virus" >is a piece of software which transmits itself from machine to machine without >the knowledge or permission of either a user on the system or the system >administrator of the machine. While I agree with the first part, I must disagree with the second. A virus is nothing more than a propagating program. "Knowledge or permission" has nothing to do with the purpose of a virus. The only factor that is necessary is some sort of rules base to maximise the probability of viable propagation. Personally, I deplore the common use of viruses primarily because it is inherently destructive whether or not the programmer was intentionally malicious. The current crop of PC viruses (what most people know as viruses is a function of personal computers - single tasking unprotected architectures) is obviously only a subset of Dr. Cohen's envelope. The incredible diversity of what the world considers a "PC" is what makes even the most innocuous virus destructive in some cases. Take STONED for example. It has only two functions: 1) To propagate 2) To occasionally display a message. The fact that it (and its close variants) are statistically the most common virus in the world today indicates that it is very good at (1). However, in some cases, probably not understood by its creator, STONED is destructive. Hard disks created without any hidden sectors (early FDISK), floppy disks with nearly full root directories, and UNIX systems may become unusable. This type of problem also occurs with professional software and any reader can name major products that would not run on a particular machine. (Years ago the true test of a "100% compatible" PC was whether or not it could run "Flight Simulator" properly. The interesting thing about FS was that the early versions ran without any operating system, you just booted the PC with the FS disk in "A:"). The point that I am trying to make is that very few people really understand PC architectures at the BIOS/Microcode level and this is necessary to be able to write "safe" low-level code. Most viruses are not intentionally destructive, however their mistakes often have the same effect. Consequently, while I can conceive of a "benevolent" virus, I would not necessarily trust one on my systems. Having said that, consider the following case: a LAN server that as part of the logon script checks the client for the presence of resident security software, verifies its integrity, and automatically updates the software on the client if missing or an older version. This would meet the test of software that is self-propagating and rules based. Even if user intervention is required to continue, given the alternative of being denied access to the LAN, few will refuse. Is this a "benevolent" virus ? (can give commercial examples). Padgett
In RISKS-13.50, I reported an article concerning software errors in auto-pilots of Boeings flown by British Airways, which appeared on the front page of the ``Sunday Telegraph'', May 17. My reason was to bring attention to the article's content, which was that there were ``10 serious incidents involving computer errors in January'' with BA. I then made a number of comments, principally that this appeared to be a high incidence rate; that the errors occurred in auto-pilots which I assumed to be relatively simple systems (as compared to fly-by-wire) in which there is much experience of design; that a comment made by a British Airways spokesman that the software was CAA approved and tested for 100 hours before entering service hardly reassuring. Imagine my surprise when I received a phone call a week later from an exasperated Dan Hawkes of the CAA. I am reporting this more than a week after the fact, largely from memory. His main complaint was that the article had been quoted without question, and that so often (as we know from newspaper reporting of our own fields) these articles are of dubious reliability and sensational. He made a further comment that he felt that academic input to the issue of software reliability in aircraft was largely negative. He reported to me that the software problems in the auto-pilots arose as a result of a modification to software; the cause had been rapidly located and fixed. Recovering from the initial shock of his call, I attempted to don a journalistic hat and ask a number of questions. I suggested that the MTBF of 10^-9 for software is unverifiable. This he was happy to agree with, but stated that auditing and monitoring of all stages of the software design and development gave a high level of confidence in its performance. Overall design meant that no single possible on-board failure (be it software of mechanical) could result in loss of aircraft integrity. He stated that as all of these involved auto-pilots, there was never any danger to the aircraft as pilots are always there to take remedial action when necessary. In effect, that these were not serious errors at all. I think Nancy Leveson (a name he was familiar with - ``an academic'') has pointed out the dangers of making highly trained pilots into computer monitors. I then raised the point that this certainly cannot apply to fly-by-wire software, as in this situation pilots are not monitors but dependent users. His answer was that the auditing and monitoring is more rigorous in the design and development of fly-by-wire, and that (to paraphrase) ``there have not been any failures yet''. Again his message was re-assurance; there is no serious risk. I could not get a real answer as to where the 10^-9 figure came from. I then decided to attempt to get in contact with the authors of the original article, Robert Matthews and Christopher Elliot. Robert Matthews (Science Correspondent) told me that the basis of the article had come from Flywise (as pointed out by Martyn Thomas, RISKS-13.51), and had been checked out with BALPA (union), BA and CAA (who were ``not all that helpful'') before publication. He stood by the article, and added that the airline companies and authorities were a closed world, and getting any information from them near impossible. Sounds familiar? He had not received any satisfactory explanation of the software reliability figure of 10^-9. I swapped sources; a few issues of RISKS for a few tidbits from him. The issue of Flywise states that the software incidents were due to ``software design defect[s]''. An interesting titbit was a paper from Boeing on structural airworthiness. According to their figures, in terms of hull loss rates per departures, to 1988 the A320 was worse than any other commercial jet since the Comet. Though none due to software; that hasn't happened yet. Simon Marshall, Dept. of Computer Science, University of Hull, Hull HU6 7RX, UK Email: S.Marshall@Hull.ac.uk Phone: +44 482 465181 Fax: 466666
Please report problems with the web pages to the maintainer