The Apache License (v2) has a dangerous interpretation

Edit: Also, the fact that the interpretation can exist and is in written form is what suddenly makes it an issue, at the very least for anyone going forward. (IANAL)

The language limits itself to cases where a patent is necessarily infringed by combining the contribution with the product. The scenario in the guy's PDF (also--PDF?) is that an infringement is created later that has nothing to do with his contribution.

Also, as others said, 1) if Apache wrote the agreement and says that others' future contributions can't give them a license to other patents, and indications are that people rely on that interpretation, it doesn't seem like a court would read it otherwise, and 2) generally, courts aren't dumb, and they're going to look at the purpose and context of the agreement and realize this is essentially about ensuring you can't taint a project with your invention and then sue for patent infringement, and avoid reading in broader meanings.

Going a little further, maybe too far, the paragraph is in the present tense--"you hereby grant," etc. Nothing in there about your future work like you might find in, say, the IP part of an employment contract.

Given that, I have a hard time seeing how your contribution in Sep 2013 binds you to license an idea that, since it didn't exist yet and nobody talked about future inventions, neither you nor the other party could've been contemplating.

(On the other hand, if there was an idea embodied in your Sep 2013 contribution that merely wasn't patented or assigned to you yet, both parties could have been considering potential patents of the idea when reading the language. So Apache's answer A2 is justifiable.)

Finally, I have little sympathy with people's desires to be able to enforce patents against open-source projects anyway, so yeah. While the language is friendly enough to patentholders, I almost wish it weren't. :)

My statement of "non-issue in practice" was based on being an open source lawyer and having talked with a large number of companies who use Apache based CLA's (including the one i work for, though i'm not speaking for them here) over the years.

While there are questions about "claims that later become licensable", i have not seen any lawyers question the scope of the grant and claim it would cover patents implicated by the work except as it existed at time of contribution (and folks are generally aware of the 'ambiguity' that exists here). There are good legal reasons to believe the interpretation would be "as the work existed at time of contribution", and no good legal reasons i'm aware of to believe otherwise.

If it actually goes bad at some point, sure, it should be modified. There are no indications it's worth doing any more than wordsmithing any other part of the agreement.

Hence, my view: This is not a problem in practice.

Here is a direct quote from the paper from which this blog post was made:

    Also, software should not be patentable.18 19 20 21 22
    The practice should be abolished immediately for the good
    of mankind and whomsoever disagrees with me on this owes
    me $1 billion in royalties because I patented the concept
    of being an asshole.

EDIT: Mirrors:

Paper: http://cdn.taoeffect.com/other/nupic/index.html#conclusion

Blog post: http://cdn.taoeffect.com/other/apache/index.html

This misses the point. If you want an atypically wide patent grant (covering not just patents infringed by your contributed code, but also to future contributions by other people which infringe a patent of yours), then that should be done openly and explicitly. Not by the backdoor with a clause that doesn't appear to do that but could be interpreted as doing do if you took it to its logical conclusion.

(Though it's pretty clear that doing do wasn't actually their intention, given their FAQ denies that interpretation - see DannyBee's post)

[1] http://www.gnu.org/licenses/license-list.html#X11License