I haven't seen any official source that describes the reason for blinking
disk indicator after Safely remove,
so what I write here is only based on general knowledge and observation.
This happens for some disks because of the way that the firmware
interacts with the operating system.

Windows announces that the device is safe to remove once it has flushed
all its data from the memory cache and signaled the device's firmware
to spin down.

The firmware itself may take some time to process the spin down command.
Many USB hard drives also incorporate cache memory that will be lost if the disk
is brutally powered down. The firmware will need its own time to terminate all writes
and to spin down the disk. While it is doing that, the disk's indicator will blink.

It is certainly not safe to unplug the disk and power it down while the
lights are blinking, since they indicate that the firmware is actively
doing something.
Some disks can recover from an abrupt power down by using non-volatile cache
memory, so they can pick up when the disk is powered up from where they stopped.

Beginning from Windows 10 version 1809, it is safe to unplug the disk whenever
the disk is not blinking. Safely remove is no longer strictly necessary,
although I would still recommend it.

Data is only guaranteed safe when electrical activity of external drive has stopped. Blinking shows that the external drive is still receiving electric power from the computer, therefore we cannot be sure that some data operation will not happen.

In Linux this is easy to make sure, because unmount and detach are two separate stages:

udisks --unmount /dev/sdb1 ..LED on, power ON

udisks --detach /dev/sdb ..LED off, power OFF

Windows does not have Power Off command but only Eject, which is similar to unmount, therefore in Windows in some cases it might be better to shutdown the whole computer as that guarantees that the external drive is also free of electric power.

No, it is not (completely) safe !!!

It should be (and in almost all cases it will be), but you cannot be absolutely sure in any sense.
^{Sadly life is not (always) what it should be.}

In my personal experience only a few times in several years (and specifically not with windows 10) I remember a data loss or file system corruption if removed quickly immediately after the go-ahead of the system. But it only happened with a second person ready to remove the USB and run away (immediately) after the go-ahead of the system and when writing operations were required at the very last moment.

The blinking indicates some activities in the device, or at least an hardware (not OS related) communication to the user. The blinking of some Seagate models is an error code ^[avid] communication. You may check if, with your model, that way of blinking (short or long pulses) has a special meaning. Usually 8 short pulses mean the byte zero, and 0 is the exit code of the program with no error at all...

TLDR

Premise. In general the scope of the question is so wide that there cannot be other single universally valid answers: too many models and brands may be involved. You must refer to both the manual of your HDD and the instructions of the operating system and drivers, consider the most restrictive and yet you will not be able to be absolutely safe.

Understanding what happens. The OS was reading/writing on the USB drive (HDD,SSD).
You ask to remove this drive from the system. The OS stops/refuses the new I/O operation requests from its programs, waits the ending of the current I/O operations and checks if some lock is active on the device. When the device communicates that all the I/O operations are ended the OS will release the device declaring that you can safely remove the device.

In the ideal world it should be enough.

The problem is... The OS can answer only on the base of what the device itself answered. Usually both the device firmware and the OS drivers are closed source, so you have to rely on their documentation and on the License Agreements. The problem is that both sides are often valued considering their performances. They may rely on the fact that between the go-ahead of the system and the average human reaction time needed to unplug the device there usually is enough time for the modern devices to finish the internal operations in progress.

Moreover, the responsibility of each side is limited by the License terms. Limitations written to safeguard the company as much as it is possible, even if they knew or should have known about the possibility of the damages (see below for Windows 10, but you can find similar for the devices producers).

So, since they discharge the consequences in advance and are valuated on their performances, it is up to you to decide to what extent you can trust them and you will feel safe.

When you have at least one side that is open source you can read what happens

^{From a comment of another answer.}

It returns when the drive claims it has flushed its own cache to the platters. You can read the SQLite source code or some NSFW comments in the Linux kernel source code about drives that are, shall we say, "less than honest" in the interest of improving benchmark figures. –

Safe and its meanings.

Now we can even specify in which sense you want to be safe...

• Data transfer safe. As I really rarely experienced, you may believe to have saved your data on the external drive, but it was not all saved, or the FAT (or equivalent) was not updated... and next time you check you will find some sector to fix or that some files are not present. Write disk cache enabled may had played its role in the past.




• Hardware safe. On some external HDD drive you can feel (and I did) that for some seconds after the system said you can remove, they still spin (you can even hear the vibrations). Unplugging the cable you interrupt the energy supply. If the plates are not spinning at their maximum speed it can rarely occur that there is not enough energy to park the heads safely onto the parking ramp ^[2].




• Privacy safe. Ok let's suppose that the firmware is moving some data from a location to another (Some SSD do this in background to rotate the sectors used, some HDD firmware may do the same on sector that presented a reading problem on a sector). Before it copies the data in the new sector, then updates the FAT (or equivalent), finally frees the old sector or if required deletes the data from it. If the process is interrupted before the end, you can finish with a sector in which that data is copied, but it is not reported and managed correctly. If that was sensible data you are exposed to a security risk.




• Legally/economically safe. You are protected for the data loss up to what fixed in the License terms, after that you will be able to proof the guilt and the culprit...

Conclusion: Wait some second more. :-)

From Win10 License Term [9.d]^[3]

The damage exclusions and remedy limitations in this agreement apply even if repair, replacement, or a refund does not fully compensate you for any losses, if Microsoft, or the device manufacturer or installer, knew or should have known about the possibility of the damages, or if the remedy fails of its essential purpose.

moreover

If your local law allows you to recover damages from Microsoft, or the device manufacturer or installer, even though this agreement does not, you cannot recover more than you paid for the software (or up to $50 USD if you acquired the software for no charge).

To your precise question, no, it is not safe until the blinking stops. Normally I do not see any blinking, but sometimes (not often) I do and wait. That works.

I would consider it "not safe" for the simple reason that you have absolutely no idea what is going on inside a drive nowadays (nor does Windows).

Windows may send a flush command and get a "Yeah OK" reply, and you pull the plug, only to discover the next time you plug in the disk that nothing was OK.

So... the drive blinks for 10 second before powering down, fine. It will take 10 seconds extra, really... what do those 10 seconds cost you? What are these seconds compared to losing data? Nothing! Just give it those 10 seconds.

Modern drives (virtually all external Seagate drives, in particular) do weird stuff such as shingled magnetic recording which increases density by overlapping tracks, and causing massive write amplification in some non-obvious conditions. You write one byte, and the harddisk may have to reorder a megabyte or two (maybe?). The controller may find a checksum error and relocate the sector, which may cause another few megabytes to be rewritten, whatever.

The same is true for virtually all solid state disks. You write one byte, the controller may have to erase a complete 512k block, copy 53687091 bytes from the old block, and write your single byte. Or... something. The controller may discover that the block has reached its end of life (becoming non-eraseable) and may mark it as such, then erase a different block, and possibly it will have to do the same thing 3-4 times. Some drives have two or three layers of caching, some of which are volatile and some that are not, and some really weird, obscure wear-levelling algorithms that do some very non-obvious stuff, too.

You have absolutely no way of knowing. It could take 0.5 milliseconds or five seconds (unlikely, but still) to finish safely. It might correctly report "flush OK" or it might do it incorrectly. Like, who cares about correctness. I've had USB devices (not just disks) from "reputable" manufactorers which were so awfully standard-violating broken implementations... it's horrible.

Even Windows might be configured in a way that causes "surprises". For example, it is normally considered "safe" to unplug drives at all times as long as no writes happen. Except that isn't true at all when you or someone else using the computer e.g. turned on performance mode (which enables write cache and thus definitively moves disconnecting at haphazard times deep into the realm of "not safe").

Even more weird stuff (MAMR, and whatnot) is on its way to becoming mainstream, and we cannot know how whatever will be in a disk that you buy in 6 months from now will work (could takes 10 seconds during which it shoots microwaves at the platter to "finalize" the disk, whatever?), nor could the guys who once upon a time wrote the "eject safely" code possibly know.

The only thing that is really 100% safe is when there are no blinking lights and no sounds coming from the drive any more (sound is admittedly not such an awesome indicator on SSDs, but on a hard disk it's a pretty good one).