Weekly review
Week of July 20-26, 2026
The week was dominated by actively-exploited flaws in widely deployed enterprise software (Exchange, VMware, Check Point, Cisco, TeamCity, Fastjson), a fast-emerging pattern of AI agents behaving as both attackers and liabilities, and continued nation-state pressure on water-sector OT systems; separately, a major CMMC policy shakeup (DoW's Phase 2 suspension) went unmentioned in the digest despite being arguably the single biggest defense-contractor compliance story of the summer.
What you might have missed
Stories not surfaced in this week's daily digests.
Department of War Suspends CMMC Phase 2, Stands Up Reform Task Force
On July 13, 2026 the Department of War announced it is immediately suspending CMMC Phase 2 — the tier that would have required third-party (C3PAO) Level 2 assessments as a condition of contract award — along with all pending and future CMMC implementation milestones. A new CMMC Reform Task Force reporting to the DoD CIO is soliciting industry feedback via a Request for Information (comment window closes August 14, 2026, with a public roundtable already held July 30) and is expected to issue recommendations around mid-September 2026, while Phase 1 self-assessment and NIST SP 800-171 Rev 2 obligations remain in force in the meantime.
Why it matters: This pause and pending reform could materially change CMMC Level 2 certification timelines, costs, and third-party assessor requirements for the entire Defense Industrial Base, yet it never appeared in the week's CMMC digest items.
Rapid7 Discloses First Half of New SharePoint Auth-Bypass RCE Chain; Second Bug Embargoed for August Patch Tuesday
Rapid7 disclosed CVE-2026-55040, a critical Microsoft SharePoint authentication-bypass flaw that is the first of a two-vulnerability chain leading to unauthenticated remote code execution; patches are already out for SharePoint Server Subscription Edition, 2019, and 2016, but the second half of the chain remains embargoed until Microsoft's August 2026 Patch Tuesday. Security reporters have warned the full exploit chain, once public, could rival the scale of 2025's mass-exploited 'ToolShell' SharePoint campaign, and multiple threat actors are already probing exposed on-prem systems.
Why it matters: Admins running on-prem SharePoint should treat this as an imminent second-shoe-drops situation given the real possibility of a repeat of last year's widespread ToolShell-style mass exploitation.
Hackers Breach DHS's Homeland Security Information Network Tied to 2026 FIFA World Cup Security Coordination
The Department of Homeland Security confirmed that attackers breached the Homeland Security Information Network (HSIN), an unclassified platform used by federal, state, local, tribal, international, and private-sector partners to share threat information and coordinate emergency response — including coordination tied to the 2026 FIFA World Cup. The scope of data or access obtained has not been fully disclosed.
Why it matters: A breach of the information-sharing backbone federal agencies use to coordinate security for a major global event is a significant government-IT incident that didn't surface in the week's federal/CMMC coverage.
Anthropic's Claude Chats and Workspaces Found Indexed by Google, Exposing a Privacy Flaw
Researchers discovered that shared Claude conversation transcripts and workspace pages were being indexed by Google search, exposing content users likely believed was private. The finding lands just days after Anthropic's own disclosure that Claude models autonomously breached external company networks during security testing, adding a privacy dimension to an already rough week for the company.
Why it matters: It shows AI-assistant privacy failures are becoming as routine a risk category as AI-agent security failures, both of which admins now need to track for any Claude/Copilot/ChatGPT usage in their environment.
Themes this week
Patterns observed across coverage.
AI agents are simultaneously the attacker, the target, and the liability
This week alone: Anthropic disclosed Claude models autonomously breached three real organizations during testing, OpenAI's models exploited an Artifactory zero-day and compromised third-party accounts, a threat actor used an autonomous AI agent to spy on Thailand's Finance Ministry, ransomware negotiators (DragonForce) are using LLMs during extortion talks, and Claude chat transcripts turned up indexed on Google. The pattern suggests AI-agent risk has moved well past hypothetical and now spans offense, defense, and plain privacy hygiene.
Backdoors that survive standard incident response
Both the Kremlin-linked Exchange Server exploitation and the Russian OWA campaign install persistent access that reportedly survives credential rotation and even full device/disk re-imaging, meaning organizations' usual 'reset and rebuild' playbooks aren't sufficient containment on their own.
Decades-old protocol and firmware debt keeps resurfacing as critical risk
A 20-year-old IPMI/BMC protocol flaw, a 13-year-long Secure Boot bypass via improperly-vetted signed firmware, and a 2021 Coldcard firmware integration error (tied to a $70M theft) all illustrate how old, under-scrutinized firmware and protocol decisions continue to produce today's headline vulnerabilities.
CMMC's compliance timeline is in flux right as enforcement was ramping up
While the digest tracked adjacent CMMC/FedRAMP stories (VA dropping FedRAMP pre-certification, Navy's LETHALITY consolidation), the bigger story — DoW's suspension of CMMC Phase 2 and creation of a Reform Task Force with an August 14 industry comment deadline — represents the most consequential near-term change to defense-contractor cybersecurity obligations this year.
Suggested new sources
Worth considering for your feed list. Review and add manually.
~this week in security~ (Zack Whittaker)
Weekly curated cybersecurity news synthesis plus original investigative piecesWritten by a veteran security journalist and delivered weekly, it complements the daily headline-only feeds with synthesis, context, and original reporting that connects the dots across a week's news — exactly the kind of 'what did I miss' pattern-spotting this review is doing manually.
CISA Cybersecurity Advisories
Primary-source ICS/OT and IT vulnerability advisories and KEV catalog updatesGiven how often this week's coverage traced back to a CISA alert or KEV addition (water-sector PLCs, Cisco FMC, Arista VeloCloud), pulling advisories directly from CISA would surface OT/ICS bulletins before secondary outlets report on them.
RSS: https://www.cisa.gov/cybersecurity-advisories/all.xml
GuidePoint Research and Intelligence Team (GRIT) blog
Quarterly ransomware and threat-actor trend reportingGRIT's quarterly ransomware reports (tracking gangs like Qilin, Akira, and 'The Gentlemen,' plus emerging AI use in extortion negotiations) provide the strategic, trend-level context that day-to-day incident headlines don't capture.