Monthly
Concurrency slot exhaustion in Pydantic AI 2.10.0 through 2.52.x lets streamed model requests permanently occupy slots in a shared ConcurrencyLimiter, progressively starving later requests until a service that reuses one long-lived limiter can no longer make model calls - an availability-only denial of service (C:N/I:N/A:H) with no confidentiality or integrity impact. The issue is remotely reachable and unauthenticated (PR:N), but our independent assessment rates it AC:H rather than GitHub's AC:L because exploitation requires a specific deployment pattern: model requests wrapped in ConcurrencyLimitedModel (or covered by limit_model_concurrency) that share a single long-lived limiter across streamed responses; agent-level max_concurrency and all non-streaming model requests are explicitly unaffected, and a short-lived or per-request limiter never accumulates enough leaked slots to deadlock. Slots are leaked by early stream termination, cancellation, consumer exceptions, or even full stream_text() consumption with debounce_by=0.1, and the vendor-released fix in 2.53.0 is available; no public exploit code identified at time of analysis.
Denial of service in Apache Tomcat's asynchronous WebSocket write path allows remote, unauthenticated attackers to exhaust server resources by triggering lost write timeouts, so non-blocking WebSocket sends are never released after their effective lifetime. All maintained branches are affected - 11.0.0-M1 through 11.0.25, 10.1.0-M1 through 10.1.59, and 9.0.0.M1 through 9.0.121, plus the end-of-life 8.5.0-8.5.100 and 7.0.43-7.0.109 lines - although exposure is bounded to instances whose deployed applications actually perform asynchronous (non-blocking) WebSocket writes; servers with no WebSocket endpoints or only synchronous writes are not exposed. The flaw is rated CVSS 7.5 (AV:N/AC:L/PR:N/UI:N, availability-only) and CISA's SSVC assessment records no observed exploitation, with no public exploit code identified at time of analysis; the primary risk is sustained resource pressure that degrades or halts the affected connector rather than data compromise.
Memory exhaustion in InternLM LMDeploy through 0.17.0 can be triggered by unauthenticated remote attackers who send repeated completion requests to the proxy endpoint when the system runs in DistServe prefill/decode disaggregation mode. The proxy uses user-facing session IDs instead of internal scheduler keys, so scheduler sessions are never released, causing unbounded metadata and memory accumulation that eventually OOM-kills the prefill worker. This is a scope-limited availability threat-only the non-default disaggregated topology is affected, no data is exposed or corrupted, and no public exploit code has been identified at the time of analysis.
Unauthenticated, remote denial of service affecting Cisco Secure Firewall Adaptive Security Appliance (ASA) and Threat Defense (FTD) software on 3100 Series and 4200 Series appliances, where a crafted stream of DTLS traffic causes improper resource management and forces the device to reload. The CVSS 3.1 base score is 8.6 (AV:N/AC:L/PR:N/UI:N/S:C/C:N/I:N/A:H), reflecting that no credentials or user interaction are required and that the impact is purely availability, though the scope change acknowledges the disruption extends beyond the vulnerable component itself. At time of analysis no public exploit code has been identified and the issue is not listed in CISA KEV, so exploitation requires either opportunistic scanning of exposed DTLS listeners or targeted traffic against internet-facing firewalls.
Worker process exhaustion in Erlang/OTP's inets httpd enables unauthenticated remote denial of service by exploiting a bare-catch error-handling flaw in the chunked transfer-encoding decoder, permanently consuming HTTP worker processes until the pool is exhausted and the server becomes unresponsive to legitimate clients. The defect lies in httpd_request_handler:handle_info/2, where a throw exception raised by an invalid chunk-size is silently coerced into a normal return tuple rather than propagated as an error, leaving the worker suspended indefinitely with the request timeout already cancelled and no default byte-rate watchdog to reclaim it. Affected OTP versions span three active release lines (27.x, 28.x, 29.x) in their default configurations; no public exploit is confirmed, though the fix commit includes a regression test that is functionally equivalent to a proof-of-concept attack.
Erlang/OTP's inets HTTP server (httpd) is vulnerable to a slow-body resource exhaustion attack exploitable by unauthenticated remote attackers across OTP versions 17.0 through 29.0. The request handler cancels its active timeout after successfully parsing request headers, then re-arms the socket for body data without installing any replacement timer, allowing an attacker to open connections with syntactically valid headers and a large Content-Length and then stall indefinitely - permanently occupying a worker process per connection. Vendor patches are available for the OTP 27.x, 28.x, and 29.x branches; no public exploit code or CISA KEV listing has been identified at time of analysis.
Incremental direct-memory exhaustion in Netty's DNS codec (netty-codec-dns) allows unauthenticated remote attackers to drive a Netty-based DNS server or DnsNameResolver into denial of service by sending DNS packets that carry malformed domain names. Because IDN.toASCII() or the domain-name encoder throws an IllegalArgumentException inside the record constructor before the associated ByteBuf is stored and later released, each crafted packet leaks a small direct-memory buffer; the leak is cumulative, so a sustained packet stream is required to exhaust memory rather than a single request. Affected releases are netty-codec-dns up to and including 4.1.135.Final and 4.2.0.Final through 4.2.15.Final; vendor patches are available in 4.1.136.Final and 4.2.16.Final, EPSS is low (0.33%, 26th percentile), and no public exploit code or confirmed active exploitation was identified at time of analysis.
Cryptographic key material in Bouncy Castle for Java FIPS (BC-FJA) persists in the JVM heap far beyond its intended lifetime on Java 12 and later runtimes because the library's zeroisation scheme relies on Object.finalize(), a mechanism progressively de-emphasized since Java 9 and deprecated for removal in Java 18. Under load, the single finalizer thread cannot retire pending objects as fast as they are allocated, causing the queue to grow without bound - resulting in an OutOfMemoryError that crashes the application and defeating the cryptographic hygiene goal of prompt key zeroisation. No public exploit is identified at time of analysis; however, any network-accessible application built on BC-FJA prior to the fixed versions is exposed to both a remotely triggerable denial-of-service and prolonged in-memory residence of sensitive key material.
Denial of service in Cisco IOS XE Software's SNMP subsystem allows an authenticated remote attacker to force an affected device to reload by sending a malformed SNMP request, disrupting all traffic transiting the device. The vulnerability spans all three SNMP protocol versions (1, 2c, and 3) and affects an exceptionally broad version range across both the 3.x, 16.x, and 17.x release trains. No public exploit or CISA KEV listing has been identified at time of analysis, but the scope-change rating (S:C) reflects that a device reload cascades into a network outage for all downstream systems dependent on the affected device.
Denial of service in the Linux kernel SCSI core allows a device removal process to hang when SCSI devices in a partially-removed state (SDEV_CANCEL) retain requeued requests that are never kicked. Because scsi_run_host_queues() relied on shost_for_each_device()/scsi_device_get(), which skips devices not in a runnable state, requests queued during host error recovery stayed stuck after recovery cleared. The fix broadens scsi_run_host_queues() to run queues for every device except those in SDEV_DEL or where a reference cannot be acquired. No public exploit identified at time of analysis; EPSS is low (0.17%, 6th percentile).
Concurrency slot exhaustion in Pydantic AI 2.10.0 through 2.52.x lets streamed model requests permanently occupy slots in a shared ConcurrencyLimiter, progressively starving later requests until a service that reuses one long-lived limiter can no longer make model calls - an availability-only denial of service (C:N/I:N/A:H) with no confidentiality or integrity impact. The issue is remotely reachable and unauthenticated (PR:N), but our independent assessment rates it AC:H rather than GitHub's AC:L because exploitation requires a specific deployment pattern: model requests wrapped in ConcurrencyLimitedModel (or covered by limit_model_concurrency) that share a single long-lived limiter across streamed responses; agent-level max_concurrency and all non-streaming model requests are explicitly unaffected, and a short-lived or per-request limiter never accumulates enough leaked slots to deadlock. Slots are leaked by early stream termination, cancellation, consumer exceptions, or even full stream_text() consumption with debounce_by=0.1, and the vendor-released fix in 2.53.0 is available; no public exploit code identified at time of analysis.
Denial of service in Apache Tomcat's asynchronous WebSocket write path allows remote, unauthenticated attackers to exhaust server resources by triggering lost write timeouts, so non-blocking WebSocket sends are never released after their effective lifetime. All maintained branches are affected - 11.0.0-M1 through 11.0.25, 10.1.0-M1 through 10.1.59, and 9.0.0.M1 through 9.0.121, plus the end-of-life 8.5.0-8.5.100 and 7.0.43-7.0.109 lines - although exposure is bounded to instances whose deployed applications actually perform asynchronous (non-blocking) WebSocket writes; servers with no WebSocket endpoints or only synchronous writes are not exposed. The flaw is rated CVSS 7.5 (AV:N/AC:L/PR:N/UI:N, availability-only) and CISA's SSVC assessment records no observed exploitation, with no public exploit code identified at time of analysis; the primary risk is sustained resource pressure that degrades or halts the affected connector rather than data compromise.
Memory exhaustion in InternLM LMDeploy through 0.17.0 can be triggered by unauthenticated remote attackers who send repeated completion requests to the proxy endpoint when the system runs in DistServe prefill/decode disaggregation mode. The proxy uses user-facing session IDs instead of internal scheduler keys, so scheduler sessions are never released, causing unbounded metadata and memory accumulation that eventually OOM-kills the prefill worker. This is a scope-limited availability threat-only the non-default disaggregated topology is affected, no data is exposed or corrupted, and no public exploit code has been identified at the time of analysis.
Unauthenticated, remote denial of service affecting Cisco Secure Firewall Adaptive Security Appliance (ASA) and Threat Defense (FTD) software on 3100 Series and 4200 Series appliances, where a crafted stream of DTLS traffic causes improper resource management and forces the device to reload. The CVSS 3.1 base score is 8.6 (AV:N/AC:L/PR:N/UI:N/S:C/C:N/I:N/A:H), reflecting that no credentials or user interaction are required and that the impact is purely availability, though the scope change acknowledges the disruption extends beyond the vulnerable component itself. At time of analysis no public exploit code has been identified and the issue is not listed in CISA KEV, so exploitation requires either opportunistic scanning of exposed DTLS listeners or targeted traffic against internet-facing firewalls.
Worker process exhaustion in Erlang/OTP's inets httpd enables unauthenticated remote denial of service by exploiting a bare-catch error-handling flaw in the chunked transfer-encoding decoder, permanently consuming HTTP worker processes until the pool is exhausted and the server becomes unresponsive to legitimate clients. The defect lies in httpd_request_handler:handle_info/2, where a throw exception raised by an invalid chunk-size is silently coerced into a normal return tuple rather than propagated as an error, leaving the worker suspended indefinitely with the request timeout already cancelled and no default byte-rate watchdog to reclaim it. Affected OTP versions span three active release lines (27.x, 28.x, 29.x) in their default configurations; no public exploit is confirmed, though the fix commit includes a regression test that is functionally equivalent to a proof-of-concept attack.
Erlang/OTP's inets HTTP server (httpd) is vulnerable to a slow-body resource exhaustion attack exploitable by unauthenticated remote attackers across OTP versions 17.0 through 29.0. The request handler cancels its active timeout after successfully parsing request headers, then re-arms the socket for body data without installing any replacement timer, allowing an attacker to open connections with syntactically valid headers and a large Content-Length and then stall indefinitely - permanently occupying a worker process per connection. Vendor patches are available for the OTP 27.x, 28.x, and 29.x branches; no public exploit code or CISA KEV listing has been identified at time of analysis.
Incremental direct-memory exhaustion in Netty's DNS codec (netty-codec-dns) allows unauthenticated remote attackers to drive a Netty-based DNS server or DnsNameResolver into denial of service by sending DNS packets that carry malformed domain names. Because IDN.toASCII() or the domain-name encoder throws an IllegalArgumentException inside the record constructor before the associated ByteBuf is stored and later released, each crafted packet leaks a small direct-memory buffer; the leak is cumulative, so a sustained packet stream is required to exhaust memory rather than a single request. Affected releases are netty-codec-dns up to and including 4.1.135.Final and 4.2.0.Final through 4.2.15.Final; vendor patches are available in 4.1.136.Final and 4.2.16.Final, EPSS is low (0.33%, 26th percentile), and no public exploit code or confirmed active exploitation was identified at time of analysis.
Cryptographic key material in Bouncy Castle for Java FIPS (BC-FJA) persists in the JVM heap far beyond its intended lifetime on Java 12 and later runtimes because the library's zeroisation scheme relies on Object.finalize(), a mechanism progressively de-emphasized since Java 9 and deprecated for removal in Java 18. Under load, the single finalizer thread cannot retire pending objects as fast as they are allocated, causing the queue to grow without bound - resulting in an OutOfMemoryError that crashes the application and defeating the cryptographic hygiene goal of prompt key zeroisation. No public exploit is identified at time of analysis; however, any network-accessible application built on BC-FJA prior to the fixed versions is exposed to both a remotely triggerable denial-of-service and prolonged in-memory residence of sensitive key material.
Denial of service in Cisco IOS XE Software's SNMP subsystem allows an authenticated remote attacker to force an affected device to reload by sending a malformed SNMP request, disrupting all traffic transiting the device. The vulnerability spans all three SNMP protocol versions (1, 2c, and 3) and affects an exceptionally broad version range across both the 3.x, 16.x, and 17.x release trains. No public exploit or CISA KEV listing has been identified at time of analysis, but the scope-change rating (S:C) reflects that a device reload cascades into a network outage for all downstream systems dependent on the affected device.
Denial of service in the Linux kernel SCSI core allows a device removal process to hang when SCSI devices in a partially-removed state (SDEV_CANCEL) retain requeued requests that are never kicked. Because scsi_run_host_queues() relied on shost_for_each_device()/scsi_device_get(), which skips devices not in a runnable state, requests queued during host error recovery stayed stuck after recovery cleared. The fix broadens scsi_run_host_queues() to run queues for every device except those in SDEV_DEL or where a reference cannot be acquired. No public exploit identified at time of analysis; EPSS is low (0.17%, 6th percentile).