Interestingly--after some outcry--this Amnesty report was submitted to an external review by a handful of academics. As with all things related to interpretation of international law, this is hardly the ultimate word, but interesting to read some other takes on the law angle. In particular, some of that relates to Dale's point "as far as possible":
The Panel notes that the duty of a defender to issue warnings is not expressly provided for by Article 58 AP I, in contrast to the express duty in this regard imposed on the attacker by Article 57(2)(c) AP I. That said, the Panel understands that AI takes a progressive approach to Article 58 AP I, interpreting it to specifically include a duty to warn in the general obligation to take other necessary precautions. If this approach is taken, conclusions could be reasonable, at least in those situations in which the civilian population was genuinely unaware that they had been exposed to an elevated
risk of attack by the adversary because defending forces were located nearby. The PR, however, should have stated more clearly that it was adopting a progressive
interpretation of IHL.
The report quotes article 58 in full:
The Parties to the conflict shall, to the maximum extent feasible:
(a) without prejudice to Article 49 of the Fourth Convention, endeavour to remove the civilian population, individual civilians and civilian objects under their control from the vicinity of military objectives;
(b) avoid locating military objectives within or near densely populated areas;
(c) take the other necessary precautions to protect the civilian population, individual civilians and civilian objects under their control against the dangers resulting from military operations.
And me reading 57(2)(c), that also has an (necessity-based) escape clause:
(c) effective advance warning shall be given of attacks which may affect the civilian population, unless circumstances do not permit.
Anyhow, the Panel concludes from their reading of article 51 (of AP I) that not removing the population is not the same as using human shields:
Intermingling combatants with civilians with the intent to avoid an attack directed against the former thus constitutes a use of human shields, which is absolutely prohibited by Article 51(7) AP I. By contrast, establishing a failure to take passive precautions to the maximum extent feasible does not require or imply the existence of any intent to engage in shielding.
It also notes however that Amnesty also apparently/implicitly made this distinction, and didn't claim Ukraine was using civilians as human shields.
As for forces locations, the matter of knowledge, feasibility estimation and intent comes into play
Even accepting that other locations existed from which Ukrainian forces could have
operated, the Panel believes AI [Amnesty] could still not definitively conclude that, as a matter of law, the Ukrainian military had failed to take precautions to the maximum extent feasible. While the PR and the supporting data reviewed by this Panel indicate that the researchers did consider feasibility in their work, the only evidence that Ukrainian forces could have located in other,
equally beneficial places more removed from civilians is the opinion of the researchers themselves. Amnesty did not have information from the Ukrainian military concerning why its forces located in the positions that they did.
In the Panel's view, there are a number of potential military considerations that might have
supported an assessment by Ukrainian forces that alternative positioning was not feasible. [...] Such considerations
might have included, inter alia, the need to position artillery in a specific location to facilitate
a necessary angle of attack on Russian forces; establishing positions that facilitated line of sight
connectivity for radar or other communication capabilities or for monitoring enemy positions
or forces or gaining access to certain infrastructure such as power or lines of communication
necessary to facilitate military operations. Similarly , the operational needs of military units
might have required them to billet in a sheltered place with electricity, access to sanitation and
potable water, and easy access to roads. According to, its researchers did take at least some of these considerations into account. However, there is no reference to them in the in any other piece of written analysis.
So, while they don't explicitly talk about the exact scenario I had in my initial Q (that an ambush in an urban environment may be a good way to stop a breakthrough), clearly the Panel considered that easing the job of the defense in some [other] ways (access to roads and what not) can be considered a factor that renders alternative placements of units infeasible.
By the by, the Panel notes (from reviewing the AI preparatory works) that there was internal dissent within AI on this matter from the original PR conclusions.
And relating to evacuation of civilians:
While it was appropriate to refer to the possibility of evacuations, the Panel believes that framed the nature of Ukraine's obligations too categorically. Belligerents are not under an absolute obligation to evacuate civilians. Instead, they should "endeavour" to do so . Like other passive precautions, evacuations are to be undertaken "to the maximum extent feasible".
and very significantly, evacuations might not have been feasible or advisable from a
humanitarian perspective. There is no indication of what alternative arrangements for civilians
existed or whether forcing them to leave their homes might not have put them in an even more
vulnerable situation. There is also no indication in the record of what the specific civilians
concerned actually wanted, in particular whether they wanted to leave their homes.
So, the Panel's view is that the presence of civilians can't be faulted to the defender, unless additional investigation takes place to know whether they wanted to move out, or if moving them was likely to immediately improve their situation in those specific circumstances.
And because one answer here talked about hospitals, interestingly enough there's a bit on that too in the review.
A section of the PR is entitled "Military bases in hospitals". In this section, reports that its researchers witnessed Ukrainian forces using hospitals as de facto military bases in five locations. In two towns, dozens of soldiers were resting, milling about, and eating meals in hospitals. In another town, soldiers were firing from near the hospital.
After quoting Article 19 GC IV, which requires warning to be given when attacking such positions,
[...] The Panel considers that, although particularly concerning from a humanitarian point of view, such co-location does not raise different legal problems under IHL than the other cases of co-location.
In other words, they [quite interestingly] seem to conclude that placing your troops in hospitals is not an outright violation of the laws of war. And in case there's doubt, they repeat in the next para that the rules
"do not explicitly prohibit such conduct by the defender".
The review then elaborates that only placing combat units next to medical units is prohibited, to a good extent (no shielding allowed), but not totally ("subject to feasibility"):
Article 12(4) AP I states: "Under no circumstances shall medical units be used in an attempt to shield military objectives from attack." AI does not claim in the PR that Ukrainian forces had this intent necessary for shielding. Beyond this absolute prohibition, the prohibition of co-location of forces and medical establishments is subject to feasibility. Article 12(4) AP I goes on to require that "[w]henever possible, the Parties to the conflict shall ensure that medical units are so sited that attacks against military objectives do not imperil their safety". The wording of Article 18(5) GC IV is even weaker : In view of the dangers to which hospitals may be exposed by being close to military objectives , it is recommended that such hospitals be situated as far as possible from such objectives". Mere presence of members of the armed forces in medical establishments does not violate these rules.
However, the Panel admits in the next para that their interpretation [in re medical establishments] is not universally accepted:
Nevertheless, AI would have been justified in arguing that any co-location by the defender violates the obligation of all parties to protect medical units, foreseen in Article 18(1) GC IV, Article 12(1) AP I, and Rule 28 of the ICRC Customary Law Study. The commentary to Rule 28 states that "[s]ome military manuals stipulate that medical units may not be used for military
purposes or to commit acts harmful to the enemy. Other manuals consider that the improper use of privileged buildings for military purposes is a war crime." Such practice must also be taken into account in the interpretation of the treaty rules requiring parties to protect hospitals, which concern both enemy hospitals and a party's own hospitals . It seems logical that if certain conduct may lead to the loss of protection of hospitals from attacks and a defending party has
an obligation to protect hospitals, it may not engage in co-location itself.
So, while the review Panel did not agree with AI's interpretation, they concluded nonetheless that AI's interpretation fell within the "reasonable" range of IHL interpretations in that regard [--hospitals].
There's also a section about schools in the review, but I suspect you may guess what the panel was going to say about that given the kind of leeway they give to the defense regarding hospitals.