AI and IT News Recap: June 29, 2026: OpenAI Gates GPT-5.6 Behind a Government Guest List, Russian Spies Hunt Signal Recovery Keys, and a Cisco SD-WAN Zero-Day Builds a Backdoor
By Noah Smith, Owner & Consultant, KeyChange Technologies · June 29, 2026
The weekend's AI and IT news was dominated by one theme: who gets to use the most powerful tools, and at what cost. Here is your skimmable recap of the stories that matter for your business, covering Friday through Sunday, June 26 to 28, 2026.
📌 The AI and IT news at a glance
- OpenAI ships GPT-5.6 but Washington gets the guest list. The new flagship models launch in a government-vetted "trusted partners" preview, not a public release.
- The "tokenmaxxing" era is ending. Enterprises are clamping down on runaway AI bills and chasing cheaper, efficiency-first models.
- GPT-4.5 is gone from ChatGPT. The older model retired on June 26; old chats roll over to GPT-5.5.
- FBI and CISA: Russian spies are now phishing your Signal recovery key. A single pasted code can hand over your entire message history.
- A Cisco SD-WAN zero-day built itself a hidden root account. Mandiant details how attackers backdoored network gear and erased the evidence.
- A CISA patch deadline landed Sunday. Federal agencies had until June 28 to fix an actively exploited Cisco call-server flaw.
🔝 Top story: OpenAI launches GPT-5.6, but the US government holds the guest list
On Friday, June 26, OpenAI previewed its next-generation GPT-5.6 family, three models named Sol, Terra, and Luna. Sol is pitched as the company's strongest model yet, Terra is a cheaper everyday workhorse priced at roughly half of GPT-5.5 with similar performance, and Luna is the lowest-cost option. The headline, though, was not the models. It was who can use them. OpenAI said it is complying with a US government request to start with a limited preview for a "small group of trusted partners" whose participation has been shared with the government, before any broader release.
OpenAI was visibly uncomfortable about the arrangement. The company stated it "believes in broad access" and said it is working to make the three models generally available "in the coming weeks," framing the restriction as a temporary security measure rather than a new normal. This follows the same pattern that kept rival frontier models gated earlier in June, and it cements a shift: the most capable AI is increasingly released on the government's timeline, not the market's.
In short: OpenAI launched GPT-5.6 (Sol, Terra, and Luna) on June 26 but limited access to a small set of government-approved partners at the request of the US government.
What it means for your business: If you build on frontier models, plan for the newest capabilities to arrive in gated previews first. Budget for a lag between "announced" and "available to you," and keep a fallback model in your stack so a delayed rollout does not stall your roadmap.
My take: The interesting tell here is that OpenAI pushed back publicly while still complying. That tension is the story. For everyday business users this changes little today, but it signals that access to top-tier AI is becoming a policy question as much as a procurement one. Worth watching whether "coming weeks" actually means weeks.
Source: CNBC
🤖 The "tokenmaxxing" era is fading as companies hunt for AI ROI
Also on June 26, CNBC reported that the mood around enterprise AI spending has flipped. The "tokenmaxxing" phase, where employers pushed staff to burn as many AI tokens as possible without worrying much about cost, is giving way to demands for clear return on investment, tighter controls, and cheaper alternatives. The examples are striking: the CEO of AI startup Lindy moved 100% of its traffic off Anthropic's Claude to DeepSeek and watched costs fall dramatically, while Uber reportedly imposed spending tiers starting at
The pressure is sharpened by cheaper challengers. China's Zhipu released GLM 5.2, which now sits within about one percentage point of Anthropic's Opus 4.8 on one agentic benchmark at roughly a fifth of the cost. With both OpenAI and Anthropic having filed confidentially for IPOs earlier in June, analysts warn their fastest growth may already be behind them as big customers rein in token spend.
In short: Reporting on June 26 showed enterprises shifting from maximizing AI usage to controlling cost, with some switching to cheaper open or rival models.
What it means for your business: This is the permission slip to audit your own AI spend. Cheaper models are closing the quality gap fast, so the highest-priced model is no longer the automatic default. Match the model to the task and you can cut cost without cutting capability.
My take: This is the healthiest development in AI all month. The "use it for everything" gold rush was never sustainable, and a focus on ROI is exactly how this technology becomes a durable business tool instead of a runaway line item. The practical move for owners: measure what your AI is actually delivering before you renew anything.
Source: CNBC
🤖 GPT-4.5 has officially left ChatGPT
A smaller but practical note: as of June 26, GPT-4.5 is no longer available in ChatGPT, including inside custom GPTs, completing a 30-day sunset that OpenAI announced back in late May. Existing conversations that relied on GPT-4.5 automatically continue on GPT-5.5, so nothing breaks mid-thread.
In short: GPT-4.5 was retired from ChatGPT on June 26, with existing chats migrating to GPT-5.5.
What it means for your business: If any internal workflow, custom GPT, or saved prompt was tuned specifically for GPT-4.5, test it now on GPT-5.5. Behavior usually improves, but prompts tuned to an old model sometimes need small adjustments.
My take: Routine housekeeping, but a useful reminder that AI vendors retire models on their schedule, not yours. If your business depends on a specific model's quirks, you need a habit of re-testing whenever one sunsets.
Source: OpenAI Help Center
🛡️ FBI and CISA: Russian intelligence is now phishing Signal backup recovery keys
The most important security story of the weekend is also the most personal. The FBI and CISA updated their March advisory (now numbered PSA I-062626-PSA) warning that Russian intelligence groups, tracked as UNC5792 and UNC4221, have added a dangerous new step to their Signal phishing campaign: tricking targets into handing over their Signal Backup Recovery Key. Hand it over once and the attacker can restore your backup, read your private and group message history, and take over the account. Worse, the advisory notes that even creating a fresh account on the same phone number does not fully protect you, because the old key can still be used.
The lures are everyday social engineering, not exotic hacking. The advisory prints two sample messages: one disguised as a mandatory two-factor authentication rollout, the other as an urgent "data recovery" fix for messages supposedly at risk. The campaign targets high-value individuals, government and military personnel, political figures, and journalists, but the technique generalizes to anyone. The fix is blunt: generate a new recovery key in Settings, which kills the old one for future backup downloads.
In short: The FBI and CISA warned on June 26 that Russian intelligence is phishing Signal users for their Backup Recovery Keys to read message history and hijack accounts.
What it means for your business: Secure messaging is only as strong as the human holding the recovery key. Tell your team plainly: no legitimate service will ever ask you to paste a recovery key, backup code, or one-time code into a chat. Treat any "urgent verification" message as hostile until proven otherwise.
My take: This one matters because it sidesteps Signal's encryption entirely and goes straight at the user. No software patch fixes "a person was talked into pasting their key." The defense is awareness, and it is cheap: a two-minute team reminder is worth more than most security tools against this attack.
Source: BleepingComputer
🛡️ A Cisco SD-WAN zero-day quietly built itself a root backdoor
On June 27, new details emerged from Mandiant on how attackers exploited CVE-2026-20245, a high-severity command-injection flaw in Cisco's Catalyst SD-WAN Manager (vManage) and related components. By uploading a malicious file (in one case a CSV named "evil_tenant.csv") through a tenant-upload feature, the attacker executed commands as root, then appended a new privileged user (named "troot") directly into the system's password files to create a persistent backdoor account, and finally ran scripts to erase evidence of the intrusion. Mandiant indicates the flaw was exploited as a zero-day for at least two months before it was publicly disclosed.
This is network infrastructure, the kind of gear that sits quietly at the core of an organization and rarely gets a second look. A root-level backdoor here gives an attacker control over the management plane that routes and secures traffic. Cisco has released fixed versions, and patching to a remediated release is the only reliable fix.
In short: Mandiant revealed on June 27 that attackers exploited Cisco Catalyst SD-WAN Manager zero-day CVE-2026-20245 to gain root, plant a hidden "troot" backdoor account, and wipe their tracks.
What it means for your business: If you or your IT provider run Cisco SD-WAN, confirm you are on a patched release now, and check for unexpected privileged accounts. Infrastructure compromises are easy to miss precisely because the device keeps working normally while an intruder has the keys.
My take: The "erase the evidence" step is what makes this nasty. Attackers are increasingly patient and tidy, which means absence of obvious symptoms is not proof you are clean. If you own network gear from any vendor, a periodic review of admin accounts should be a standing habit, not a post-incident scramble.
Source: BleepingComputer
🛡️ A federal patch deadline for a Cisco call-server flaw landed Sunday
Rounding out the security front, CISA set Sunday, June 28 as the deadline for federal agencies to patch CVE-2026-20230, an actively exploited server-side request forgery (SSRF) vulnerability in Cisco Unified Communications Manager, the platform that runs many organizations' phone and collaboration systems. CISA added the flaw to its Known Exploited Vulnerabilities catalog the week prior, and the binding deadline forces government agencies to act fast.
Federal deadlines are not legally binding on private companies, but they are an excellent free signal. When CISA flags something as actively exploited and gives agencies a tight clock, that is a strong hint the rest of us should move quickly too.
In short: CISA required federal agencies to patch actively exploited Cisco Unified Communications Manager flaw CVE-2026-20230 by June 28.
What it means for your business: If your phone or unified-communications system runs on Cisco CUCM, treat this as your deadline as well. Ask your IT provider to confirm the patch is applied.
My take: I keep coming back to the same advice because it keeps being right: CISA's catalog is the best free prioritization list in security. You do not need to track every CVE, just the ones being exploited right now, and CISA hands you that list for free.
Source: CISA
Catch up on the previous edition: AI and IT News Recap: June 25, 2026.