scfiguess=ao produces the lowest energy at 1st step of SCF

With the following input:

# Hartree-Fock/STO-3G
mem=256MB
calc=SCF
dft=off
basis=STO-3G
verbosity=2
charge=0
mult=1
scfiguess=ao
test=-111.2245586623381683
geom=xyz
2

C  0. 0. 0.
O  0. 0. 1.128

At first SCF step, MRCC produces -113 Hartree, then, in the next steps, energy goes to -111 Hartree:

 ***HARTREE-FOCK ENERGY IN STEP   1 IS       -113.4789691892482040 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   2 IS       -111.2169101136965566 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   3 IS       -111.2219782649955562 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   4 IS       -111.2244526382469019 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   5 IS       -111.2245577846116475 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   6 IS       -111.2245585869731030 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   7 IS       -111.2245586621338163 [AU]
 ***HARTREE-FOCK ENERGY IN STEP   8 IS       -111.2245586623381683 [AU]
 ***FINAL HARTREE-FOCK ENERGY:       -111.2245586623381683 [AU]

MRCC version is mrcc.2023-08-28-release.binary taken from shared directory on Komondor.

The same problem was noticed for another system in the same basis set.
For CO system, broken basis sets are (from tested):

• STO-3G (-113.4789691892482040)
• 6-311G** (-115.4695600868956262)
• 6-311++G* (-115.4695689002730319)
• def2-SVP (-114.9122663285606336)
• def2-TZVP (Energy is too low: -94.0047953369750644)

while the following basis sets do not have such problem:

• 3-21G
• 6-31G
• cc-pVTZ
• aug-cc-pVTZ

HF limit is about -112.7906490127417385 Hartree.