At what point, given the rapidly increasing power of visible light output from LEDs, will the possibility of eye damage have to be taken seriously? I know that looking into my 1.5W LED torch it's at the point where a part of the retina will be washed out for minutes. And that's after the external lens which provides a fairly wide beam.
Some car headlamps seem particularly good at doing that ... eat more carrots !
I would gues it depends largly on the beam divergence and the stupidity of the viewer, with diverese beams of visible light the human reflex action is quite good at avoiding damage, but theres always someone wiling to stare into the beam ! of course with invisible laser light the risk is so much greater becuase you are unaware its doing damage untill its too late. however you can blind yourself if you stare at the sun. im not sure how much of this is due to infra red than visible light.
In theory, today's brighter LED could already damage the eye: the SiC LED from Cree have an optical power of 250 mW, and if you collimate the beam so that all the power enters the eye, you will have similar effects of a class IIIb LASER (definitive injury of the eye for medium-short exposures).
However, in normal situations you won't go over class I or class II. To make an example, let's use some optics so that the beam of our 250 mW LED will have a width of 30 degrees (e.g. a flashlight). If you consider a distance from the LED of 1 m and a pupil area of 10 mm^2, you will have a total beam area of 0.214 sr and a pupil area of 1e-5 sr. Therefore, the optical power that will enter into your eye would be 0.25*1e-5/0.214: less than 12 microwatt, i.e. like a class I LASER (totally harmless).
However, some time ago a Cree distributor showed me one of their LED in action, and before lighting it he warned me not to look directly at it. And he was right, these LED produce a light that you can't stand, even from several meters.
I was rather thinking of children holding a high power LED torch and shining it in their eyes (that's what I did with mine). Distance is around 30cm max (more like half that for a child), and beam angle about
15deg at a guess. At that distance the eye would receive closer to 10mW input. Next year's superbright torch might multiply that by x10.
I have seen a figure of 1kW per square meter for the intensity of sunlight at the equator on a clear day. This is 1mW/mm^2. That would mean 10mW total if the eye is 10mm^2. And, we know that sunlight has enough oomph to cause blindness. If you don't believe me, ask Euler.
Light does, but the eye has a lens in it and views a focused image. As a result, surface brightness is independent of distance. (If the light source is farther away, it covers a smaller area of the retina.) So it's a question of losing a lot of your retina or just a little...
What I mean by that is that the watts per unit area of the retina are independent of the distance of the light source. As the source gets farther away, the area of retina that it covers is what diminishes.
The risk of damage to any particular area of the retina depends on the wattage per area, not the total wattage.
I would think that if the damage is due to heating, it would be more dangerous to have a larger area illuminated at the same level. For a smaller area, conduction to the surrounding tissue and blood circulation might be able to limit the maximum temperature reached.
There is an alternative limit in terms of joules per square centimeter per steradian, which I have calculated (I hope correctly) equates to 63 joules per square centimeter of "lambertian" emitting surface during viewing period that could be 10,000 seconds for "Class I" exposure. And if I did not screw this up, surface emitting 63 joules per square centimeter (all or an adequately signidficant portion visible) in 1 to
9,999 seconds is short of Class IIIb even if total output power is over 5 mW.
I would suggest 10 degrees.
When pupil diameter is a factor with visible light, the regulation allows 7 mm pupil diameter or basically 40 mm^2 where it states relevance and where it applies.
21 CFR 1040.10
Although in practice a pupil will usually shrink to less than 10 mm^2, maybe as little as 3 mm^2 if you arev looking at "hazardously bright light".
Although this example is largely harmless, I do see the need to point out:
The upper limit of "Class I" is .4 microwatt and non-inclusive (.4 microwatt is Class II while anything less and guaranteed less is Class I no matter how little below .4 microwatt).
An LED flashlight even with a high power LED can achieve a beam area about an order of magnitude smaller - so make that 110-120 microwatts.
If you need to CYA, then you need to consider a 40 square millimeter pupil, and you could then be up against .4-.5 milliwatt.
The "acid test worst case" is a really big dark-adapted pupil or a pupil dilated by the pupil-dilating eyedrops administered sometimes by ophthalmologists, and that is 10 mm diameter, or about 79 square millimeters. But where 21 CFR 1040.10 specifies a pupil diameter where pupil diameter is relevant, it says 7 mm diameter which is about 40 square millimeters.
When it comes to LEDs and LED flashlights, you may end up relying on power per unit area of emitting surface. For example, a more intense white "Luxeon" could emit .4 watt from a chip surface magnified to roughly (my eyeball estuimate) a 1.6 mm square, and that is roughly 16 watts per square centimeter - which means you hit maximum safe exposure in about 4 seconds if you are close enough for 1/4 milliwatt or more to pass through your pupil. This is about 1/4 the upper limit of Class II.
I hear enough cautions and warnings related to looking into enough LEDs... I would consider in general high brightness LEDs to be equivalent to Class II lasers, and laser class II is a wide range of 2500 to 1.
At least 21 CFR 1040.10 and .11 do not have legal regulatory force on LEDs, incandescent lamps, halogen lamps, other non-laser light sources... Although I suspect your lawyer if competant in such area might advise to use that as a guideline for non-laser light sources.
NOTE - I am talking about visible light here. IR and UV gets worse, notably for at least some instances of Class II being changed to Class IIIb if the wavelength changes from an "officially visible" one to an "officially infrared" one. At least that does not apply to white or other high visible brightness LEDs!
A plain old ordinary dirt-cheap R-2 flashlight "lamp" ("bulb") consumes about a watt, and I would hazard to guess that about 40% of that gets radiated in "optical band" wavelengths (UVA, visible, and IRA - mostly IRA). And maybe about half that gets received and reflected by the reflector and gets out as the "main beam" - meaning somewhere around roughly .2 watt! "1-watt" "Luxeon" LEDs are hard-pressed to do that!
Let alone a "MAG" or other premium flashlight with 3, 4, or 5 cells and/or a premium "bulb" (lamp) with krypton or xenon and/or halogen enhancement - you could get half a watt or more of "optical band" radiation, difficult to exceed with 2.5 watts into a "Luxeon" LED!
UV and IR are not significant problems with white LEDs. White LEDs are pretty good at specializing in production of visible light.
Same is true of higher brightness (in photometric terms) non-white LEDs of just about any color - red, orange, yellow, green, blue, pink, pastel, "warm white", "coral", and "purple" (as opposed to violet - there is a difference, although I know of some Japanese ones said to be "purple" to actually be violet-UV).
Now for what other "electric lamps" do in terms of having "optical band" output being specialized to visible wavelengths: Fluorescent lamps and HID lamps of kinds intended for illumination do that fairly impressively well, while incandescent lamps (including halogen lamps) do not due to producing lots of "IRA band" infrared.