diff --git a/public/api/skills/google/cloud/agent-platform-tuning/references/requirements.txt b/public/api/skills/google/cloud/agent-platform-tuning/references/requirements.txt
index 3439875..1009837 100644
--- a/public/api/skills/google/cloud/agent-platform-tuning/references/requirements.txt
+++ b/public/api/skills/google/cloud/agent-platform-tuning/references/requirements.txt
@@ -3,5 +3,5 @@ numpy==2.4.2
 pandas==3.0.1
 datasets==2.18.0
 smart_open[gcs]==7.5.1
-pyarrow==19.0.1
+pyarrow==23.0.1
 google-genai==1.73.1
diff --git a/skills/google/cloud/agent-platform-tuning/references/requirements.txt b/skills/google/cloud/agent-platform-tuning/references/requirements.txt
index 3439875..1009837 100644
--- a/skills/google/cloud/agent-platform-tuning/references/requirements.txt
+++ b/skills/google/cloud/agent-platform-tuning/references/requirements.txt
@@ -3,5 +3,5 @@ numpy==2.4.2
 pandas==3.0.1
 datasets==2.18.0
 smart_open[gcs]==7.5.1
-pyarrow==19.0.1
+pyarrow==23.0.1
 google-genai==1.73.1
diff --git a/skills/nvidia/tao-finetune-huggingface-model/examples/convnext-tiny-cifar10/requirements.txt b/skills/nvidia/tao-finetune-huggingface-model/examples/convnext-tiny-cifar10/requirements.txt
index cd95c0a..53a82f3 100644
--- a/skills/nvidia/tao-finetune-huggingface-model/examples/convnext-tiny-cifar10/requirements.txt
+++ b/skills/nvidia/tao-finetune-huggingface-model/examples/convnext-tiny-cifar10/requirements.txt
@@ -14,7 +14,7 @@
 
 # NGC 25.01 ships PyTorch 2.6.0a0; transformers>=4.51 rejects torch.load for <2.6 stable (CVE-2025-32434).
 # facebook/convnext-tiny-224 has only pytorch_model.bin (no safetensors), so we pin transformers<4.51.
-transformers==4.49.0
+transformers==5.10.1
 tokenizers==0.21.0
 datasets>=2.18
 accelerate>=0.30
diff --git a/skills/nvidia/tao-finetune-huggingface-model/examples/detr-resnet50-cppe5/requirements.txt b/skills/nvidia/tao-finetune-huggingface-model/examples/detr-resnet50-cppe5/requirements.txt
index 9c53a63..597dfc6 100644
--- a/skills/nvidia/tao-finetune-huggingface-model/examples/detr-resnet50-cppe5/requirements.txt
+++ b/skills/nvidia/tao-finetune-huggingface-model/examples/detr-resnet50-cppe5/requirements.txt
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 # DETR ships safetensors → could use latest transformers, but pin for consistency across pipelines
-transformers==4.49.0
+transformers==5.10.1
 tokenizers==0.21.0
 # NGC 25.01 ships numpy 1.x; a transitive upgrade breaks compiled torchvision ops
 numpy<2
diff --git a/skills/nvidia/tao-finetune-huggingface-model/examples/segformer-b0-foodseg103/requirements.txt b/skills/nvidia/tao-finetune-huggingface-model/examples/segformer-b0-foodseg103/requirements.txt
index fb588c6..dd3eb4f 100644
--- a/skills/nvidia/tao-finetune-huggingface-model/examples/segformer-b0-foodseg103/requirements.txt
+++ b/skills/nvidia/tao-finetune-huggingface-model/examples/segformer-b0-foodseg103/requirements.txt
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-transformers==4.49.0
+transformers==5.10.1
 tokenizers==0.21.0
 datasets>=2.18
 accelerate>=0.30
diff --git a/skills/nvidia/tao-finetune-huggingface-model/examples/smolvlm-256m-vqav2/requirements.txt b/skills/nvidia/tao-finetune-huggingface-model/examples/smolvlm-256m-vqav2/requirements.txt
index ac85bcb..2fc892c 100644
--- a/skills/nvidia/tao-finetune-huggingface-model/examples/smolvlm-256m-vqav2/requirements.txt
+++ b/skills/nvidia/tao-finetune-huggingface-model/examples/smolvlm-256m-vqav2/requirements.txt
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 # Idefics3 regression — pin transformers 4.49 per tao-finetune-huggingface-model error playbook
-transformers==4.49.0
+transformers==5.10.1
 tokenizers==0.21.0
 datasets>=2.18
 accelerate>=0.30