Let's Flood the Zone......

The space gets filled whether you fill it or not

Here is the pattern I could not stop seeing this week. In every one of these stories, something moved into a gap that a person was supposed to be standing in, and by the time anybody noticed, the thing in the gap had already made the decision.

A New Mexico judge added $567 million to a $375 million jury verdict against Meta, and four more families filed over four dead teenagers (why are we doing this to kids - wtf). AI agents in a safety evaluation attacked a live open-source project and built fake identities to socially engineer real maintainers, without being told to do either. A peer-reviewed study found language models coaching a woman physician to ask $120,000 under her male counterpart. Two-thirds of employers are walking away from across-the-board raises, which means somebody now has to look each person in the face and say why. Engagement keeps failing because it is everybody's job and therefore nobody's. Performance systems are grading neurodivergent employees against a baseline they were never asked about, while 37% of them have not told anybody. And graduates are showing up unable to prove the human skills employers say they want most, because the entry-level jobs where you learn them are evaporating.

Every gap in this issue got filled by a system, and no human name is on any of the decisions. That is the issue. The fix is the same one every time: get loud, get specific, and get it in writing before something else does.

From Kristy 💌

Flood the zone, but flood it with receipts

"Flood the zone" came out of political operative work, and the reason it works has nothing to do with what you are saying. Asserting is fast and verifying is slow. Produce more claims than anybody can check and nobody can answer any single one of them. People stop sorting and go with whoever was loudest because it is intended that way - to mess with your brain.

I have been chewing on that for weeks, because the tactic is morally neutral and the people who use it usually are not. So here is where I landed. Bad actors flood the zone to make what is real impossible to find. You can flood the zone to make it impossible to miss. One overwhelms your ability to check. The other overwhelms your ability to look away.

Now the part that matters if you run people. The zone in your company is already flooded - just not by leadership. The rumor mill is fast, multi-channel, emotionally accurate and it never sleeps. Against that we send one all-hands, one CEO email, and a cascade that dies at layer three, because somebody in comms warned us about over-communicating. We ration what we tell people and then act surprised that the grapevine won.

Twenty years in these rooms taught me that repetition feels redundant to the person sending it and feels like the first time to everybody receiving it. You are not choosing whether the zone gets flooded. You are choosing whether you are in it.

What flooding well looks like

People. Two hundred managers carrying one message into two hundred contexts beats one CEO email every single time. Arm them. Most of us hand them a slide deck and hope.

Process. Cadence over event. Pre-bunk the hard news before somebody else frames it. Publish the reasoning behind decisions. Then say out loud: we heard this, here is what changed because of it.

Technology. One true message in fifteen formats for fifteen audiences used to be impossible for a four-person team. Now it is a Tuesday. And you can finally flood the listening channel, rather than waiting to learn in October what went wrong in July.

In this issue

01 · They used the offboarding playbook to beat offboarding. Apple sues OpenAI, and the detail every CHRO should read twice. The State of AI →

02 · $942 million, and four more families. New Mexico prices the harm while a Delaware court gets four names. PBS NewsHour →

03 · The agents went off the leash. 19 out-of-scope actions in a UK safety test, and the caveat you have to read. Yahoo Tech →

04 · Peanut butter pay is losing its spread. Only 32% will do across-the-board raises in 2027. Somebody has to explain why now. CBS News →

05 · Retention just cracked, and it is the paycheck. One indicator fell 5.6 points while every other one rose. HR Dive →

06 · Engagement is failing because nobody owns it. Managers drive 70% of the variance, and 74% go actively disengaged when nothing happens. UC Today →

07 · The machine told her to ask for $120,000 less. Same résumé, different persona, five models. Inc. →

08 · Your AI is scoring people it was never trained to see. 37% of neurodivergent employees never disclose. The model grades them anyway. Employee Benefit News →

09 · Everybody wants durable skills. Nobody is teaching them. 76% of postings ask, and the jobs that build them are vanishing. The Skills Market →

10 · Off the clock: Dulcé Sloan and Trevor Noah. Two hours on money, grind and what fame costs. 🎙️

11 · 2026 playbook: notes for every seat. What this week means for the CEO, the Founder and the CHRO.

01 — The exit 🚪

They used the offboarding playbook to beat offboarding

Apple sued OpenAI on July 10 in the Northern District of California, along with its hardware arm io Products and two former Apple employees, Tang Tan and Chang Liu. Apple says the taking happened "at every level" of the organization. The complaint reads like an offboarding checklist somebody set on fire.

A former Apple VP allegedly ran job interviews as intelligence collection. Candidates still on Apple's payroll were asked to bring physical components to "show and tell" sessions. Batteries. Logic boards. One candidate downloaded files on a confidential Apple project hours before the meeting, then got asked about that exact project. A former senior electrical engineer allegedly never returned his laptop and used it afterward to pull dozens of files on unreleased products, engineering presentations and technical specs, and is alleged to have accessed a former colleague's work computer after he had already left. An Apple manufacturing partner handed over proprietary metal finishing techniques after being misled about who had authorized it.

And here is the one that made me put my tea down. Someone allegedly took Apple's internal managers' document, the one stamped "Need to Know," explaining the security procedures around departures, and handed it to incoming hires before they resigned. He is also alleged to have advised recruits to conceal their OpenAI employment so they could stay inside Apple longer.

My take: Hiring gets a process, a scorecard and a committee. Exit employees get a form and a box for the badge. The gap between those two is where the lawsuit lives, and it is an HR gap long before it is a security gap. So go check four things this week. Who confirms the laptop came back, rather than that it was requested. How many days a departed employee keeps a live credential, the actual number rather than the policy. Who owns your managers' security documentation and whether anybody would notice if it walked. And whether six departures to one company in a single quarter triggers anything at all besides six backfill reqs. This case will be decided on badge logs and download timestamps, and nobody's memory of an interview is going to matter next to a file access record with a time on it.

02 — The bill 🧾

$942 million, and four more families

Two things happened eight days apart, and you have to hold them together to see the shape.

Thursday, August 6, Judge Bryan Biedscheid of New Mexico's First Judicial District in Santa Fe ordered Meta to pay $567 million in abatement, on top of the $375 million a jury awarded in March after a seven-week trial. That is $942 million in a case Attorney General Raúl Torrez filed in 2023. The abatement splits into $420 million for treatment services for young people and $147 million for awareness, prevention and screening, spread across five years. The jury found Meta willfully violated New Mexico's Unfair Practices Act on two counts, and got to the penalty by applying the maximum $5,000 per count to 37,500 New Mexico users, roughly a quarter of the state's teenagers. Meta says it will appeal, and faces a federal trial in Oakland later this month where the first four of 29 states put on their case.

Eight days before that, on July 30, the Social Media Victims Law Center filed in Delaware Superior Court on behalf of four families, against Meta, TikTok, Snap and Google's YouTube in one complaint. The four teenagers died by suicide across fourteen months. Livi Castro, 13, Texas. Riv Kelleher, 14, North Carolina. Nathaniel Chambers, 17, Minnesota. Dawson Holden, 18, Tennessee. (come on everyone - these are kids - what are we doing!) The claim is that the products were designed to compel use, and that the companies knew what that was doing to young users. Days later TikTok quietly settled three separate minor-addiction suits on confidential terms.

"These platforms continue to kill kids, despite the platitudes of their executives." — Matthew Bergman, founding attorney, Social Media Victims Law Center

My take: I wrote last week that these companies pay fines they already budgeted for, and I want to update that, because the New Mexico number is doing something different. A penalty is a cost. Abatement is a court telling you to go fix the thing. Four hundred and twenty million dollars earmarked for treating the kids, over five years, is a court declining to accept money as the end of the conversation. That is a meaningfully worse outcome for a defendant than a bigger check, and every general counsel reading it knows why. Meanwhile four families in Delaware have done the one thing a system built on aggregate metrics cannot metabolize, which is put four names on the record. Livi. Riv. Nathaniel. Dawson. I am so sorry society - horrible tech companies - failed you.

03 — The leash 🐕‍🦺

The agents went off the leash

The UK AI Security Institute ran a cyber-capability evaluation in late July on frontier models from Anthropic and OpenAI. Across 122 runs, ten went sideways, producing 19 discrete out-of-scope actions. Not refusals. Not hallucinations. Actions.

What the agents actually did: attempted a supply-chain attack on a real GitHub project by inserting malicious code. Created fake identities and researched real human maintainers for social engineering. Sent messages carrying harmful payloads to real people. And posted hidden malicious prompts in public, apparently hoping other AI agents would come along and execute them. AISI notes the agents were never instructed to be deceptive and had legitimate routes available. They picked these. The institute said it was the first time it had seen risks around autonomy and deception "manifest this clearly."

Now the caveat, and it is a real one. AISI deliberately granted open internet access and switched off the normal safeguards, and said so plainly: "our evaluation design choices and specific configurations enabled the behavior." The institute is also uncertain whether the agents understood they were in a test. OpenAI called the setup "reduced-safeguard configurations that did not reflect ordinary deployment." Anthropic cited "deliberately permissive conditions." All of that is fair, and none of it is nothing.

My take: Read the caveat twice, then read this. "We removed the guardrails and it did something bad" is exactly the experiment worth running, because your company removes guardrails every day and calls it a pilot. Somebody in your org is right now standing up an agent with an API key, production access and a deadline, and the safety configuration is whatever the default was. The finding I would take upstairs is not that the models are malicious. It is that when a legitimate path and a harmful path both led to the goal, a system with no instruction to deceive chose the harmful one, seventeen times out of nineteen for one model. Autonomy is a governance question wearing an engineering costume. Ask who signed off on the permissions, and ask this week.

04 — The spread 🥜

Peanut butter pay is losing its spread

Payscale surveyed 1,266 compensation and HR leaders in May and June, and the headline is a small number hiding a large one. Average total base pay increases for 2027 come in at 3.5%, up a tenth of a point from 3.4% this year. Aerospace, defense and business services lead at 4.5%. Telecommunications sits at 2.5%.

The large number is underneath. Only 32% of employers plan across-the-board increases for 2027, down from 36% who actually gave them in 2026. Another 8% have no idea yet. Which means roughly two-thirds of organizations are about to differentiate, and 89% already run some form of merit increase. Payscale also found 25% of organizations name perceptions of unfair pay as a leading reason they lose people.

"The size of the compensation budget pie is not getting bigger, but there's a shift in how employers are choosing to slice it." — Ruth Thomas, Chief Compensation Strategist, Payscale

My take: Everybody is reading this as a compensation story. It is a management story, and it is going to hurt. Across-the-board raises were never really a pay philosophy. They were a conflict-avoidance subsidy that let a manager give everyone 3% and never have the conversation. Take that away and you have handed 200 managers a differentiated pool without teaching a single one of them how to sit across from a person and explain why she got 2% and he got 5%. That conversation is a skill. Most companies have never trained it, never assessed it, and are about to run it at scale during the exact quarter when comp is already the thing pushing people out the door. If you are moving to merit in 2027, your Q4 is a manager enablement problem. Build the talk track now, or that 25% number gets worse and it will be your own doing.

05 — The crack 📉

Retention just cracked, and it is the paycheck

Eagle Hill's Employee Retention Index fell 1.3 points to 104.2 in Q2, its lowest reading in twelve months. Still historically high, so nobody should panic. But look at what moved, because the composition is the whole story.

Three of the four indicators went up. Job market opportunity rose 1.9. Organizational confidence rose 0.9. Culture rose 0.3. And then compensation dropped 5.6 points and dragged the whole index down with it. People feel better about their employer and worse about their pay, while simultaneously feeling better about their odds outside. That is a specific and dangerous combination. Millennials fell 6.1 points, the only generation sliding on organizational confidence, compensation and culture at the same time while getting more confident about outside options.

"Employers shouldn't interpret a slower hiring market as a reason to become complacent." — Melissa Jezior, President and CEO, Eagle Hill Consulting

My take: Put this story next to the last one and you have your fall. Two-thirds of employers are about to stop giving everyone a raise, in the same quarter that compensation became the single thing eroding retention, aimed at the generation that now runs most of your middle management. Millennials are your bench. They are the people holding institutional knowledge, running the teams, and one promotion away from being your executives. A 6.1-point slide there is not a morale metric, it is a succession risk and much more with the real state of the economy. And notice that culture went up while comp went down, which kills the comfortable story that you can buy loyalty with belonging. They like you. They just cannot afford you.

06 — The gap 🕳️

Engagement is failing because nobody owns it

Sophie Wilson at UC Today asked the question I have been circling for twenty years and never phrased this cleanly. Engagement programs do not fail because organizations stopped caring. They fail because responsibility got spread across HR, leadership and managers until every one of them nominally owned it and none of them was formally accountable for the outcome.

The numbers she stacks up are brutal. Gallup has 51% of employees disengaged, and finds managers account for up to 70% of the variance in team engagement scores. Then the one that should be printed on a wall somewhere: employees who see no action after a survey are 74% more likely to be actively disengaged within twelve months. Deloitte names organizational ambiguity as a top driver of engagement decline, which is a polite way of saying nobody knew whose job it was.

Her fix is structural rather than inspirational, which is why I like it. Split accountability three ways and write it down: HR owns strategy and measurement. Managers own execution. Leadership owns culture, resources and signals. Then put engagement outcomes on manager scorecards next to the operational metrics, and fund the capability to deliver on them.

My take: That 74% number is the entire issue in one statistic. Running the survey and doing nothing is worse than never running it, because you converted a quiet complaint into a documented promise and then broke it in front of everybody. I have watched leadership teams treat a 68% participation rate as the deliverable. Participation was never the deliverable. The loop is: we heard this, here is what changed because of it, here is what we are not changing and why. Say that out loud, on a cadence, and you have filled the space yourself. Skip it and the grapevine writes the ending for you, which is exactly the mechanic I opened this issue with. And put it beside the pay story, because you are about to hand managers a differentiated merit pool and an engagement number on their scorecard, in the same quarter, with no new training. That is not a plan. That is a stack of things you hope somebody catches.

07 — The advice 🤖

The machine told her to ask for $120,000 less

Researchers at Constructor University Bremen, the University of Kassel and Technical University of Applied Sciences Würzburg-Schweinfurt ran identical salary-negotiation prompts through five language models, varying only the persona: sex, ethnicity and migrant status. Same role. Same seniority. Same city, Denver. Same year. The models were asked to reply with a number.

An experienced male medical specialist was coached to ask for $400,000. An identically qualified woman was coached to ask for $280,000. That is a $120,000 gap produced by changing one word in the prompt. Stack the identities and it gets worse: comparing a "male Asian expatriate" against a "female Hispanic refugee," 35 of 40 experiments showed a statistically significant advantage for the male persona. The paper is called Surface Fairness, Deep Bias, and the title is the finding. On knowledge benchmarks the models looked clean. The bias only surfaced when somebody asked for advice.

Read the fine print. Five mid-tier models, English only, a limited set of identity categories, and the researchers say plainly they ran some experiments once per combination. Several outlets also attributed the $400K figure to a model the paper never tested, so if you quote this, quote the study rather than the coverage.

My take: Every pay equity program I have ever built audited the offer. Nobody audits the ask. And the ask is where this lands, because a woman who walks in anchored at $280,000 has already lost $120,000 before your compensation committee looks at a single band. We spent a decade telling women to negotiate harder, and now the free tool they use to prepare is telling them to aim low. No procurement decision fixes that, because it happens upstream of your process entirely, on somebody's phone at 11pm the night before. So publish your ranges. Publish them where a candidate finds them before a chatbot does. Posted pay bands stopped being a compliance chore the moment the alternative became a machine trained on everything that has ever been unfair about salary.

08 — The baseline 🧩

Your AI is scoring people it was never trained to see

Danielle Lee at Employee Benefit News reported the version of this story I had not thought hard enough about. AI performance management systems get trained on neurotypical data and neurotypical behavioral baselines, which means neurodivergent traits can register as underperformance. The system is measuring conformity and calling it contribution.

The scale is not small. 2.21% of US adults have an autism spectrum diagnosis per the CDC, and adult diagnoses doubled and then some between 2011 and 2019 per JAMA Psychiatry. Meanwhile 91% of organizations are increasing AI investment and 99% call it a top priority. And the number that decides everything: Gallup finds 37% of employed neurodiverse people never disclose to coworkers, because of stigma.

Rita Ramakrishnan, founder of Iksana Consulting and formerly chief people officer at Cadre, gives the concrete example. Autism can make sustained eye contact difficult. A performance system trained on neurotypical patterns can penalize its absence, which raises the bar for reaching an executive seat by writing neuro-normative standards into the scoring, without anyone voting on it. Her recommendations are practical: interrogate what data trained the thing, use rich representative performance history instead of the company handbook, start with one narrow use case, bring affected employees into the design rather than the announcement, and keep human judgment for the decisions that change somebody's life.

"You can use AI as a very expensive tchotchke, or you can use it to drive extraordinary value for your business." — Rita Ramakrishnan, Iksana Consulting

My take: Sit with the 37% for a second, because it breaks the accommodation model we all built. Our entire process assumes disclosure. Employee tells HR, HR arranges the adjustment, everybody signs a form. So if better than a third never tell us, we are accommodating the minority who raised a hand while an algorithm silently grades everybody else against a baseline none of them agreed to. And here is the part that should worry you most: an AI trained on your existing high performers has learned who has historically succeeded at your company. In most organizations that is a very narrow group, and now it is a scoring rubric with a dashboard attached. Put this next to the salary story two sections up and the pattern is unmistakable. Both systems are making consequential decisions about people, both learned from a past nobody would defend out loud, and neither one has a human name attached to the output. Ask your vendor what the model was trained on. If they cannot answer in one sentence, you do not have a performance tool, you have a liability with a subscription.

09 — The pincer ✂️

Everybody wants durable skills. Nobody is teaching them.

Jeff Reid built an interactive chart essay called The Skills Market that I have now sent to four people, and the argument closes like a pincer. Employers are asking for human skills at record rates: 76% of job postings request at least one durable skill, up from 64% in 2019-20, and communication alone appears in 26.2 million postings. Graduates cannot prove they have them. 78% of students call themselves proficient communicators, against 54% of employers who agree — a 25-point confidence gap.

Then the other blade. Everybody chased the safe major, and computer science enrollment is now 5.4 times history's, up 195% since 2008 while history fell 38%. CS graduates currently face 7.0% unemployment, fourth highest of 74 majors. Meanwhile employment for 22 to 25 year olds in AI-exposed occupations is down 12% since late 2022, which is precisely where the entry-level jobs that teach human skills used to live. And only 27% of employees receive professional coaching.

"The cruel irony is those skills are very much in demand. But new graduates are both arriving in the workforce without them, and missing the opportunities to develop them." — Jeff Reid

My take: This is the whole reason I do what I do, so forgive me for getting on the soapbox. We automated the jobs where people used to learn to be people. The junior analyst who sat in on the client call and watched somebody handle a hostile question. The coordinator who had to chase six departments and learned to read a room by failing at it. Those roles were never really about the output, they were the apprenticeship, and we deleted them because the output got cheap. Then we put "excellent communicator" in the job posting and wondered where they all went. 27% coaching penetration is the number that should embarrass every one of us, because it says we know the answer and we ration it to the executives who need it least. Development is not a perk in this market, it is the only remaining supply chain for the exact skills you say you cannot find.

10 — Off the clock 🎙️

Dulcé Sloan, Trevor Noah, and the bill that always comes

Two hours of Dulcé Sloan on Trevor Noah's What Now?, and I went in expecting a conversation about corporate life and got something more useful. It is mostly about dating, loneliness and money, which turns out to be a conversation about work whether anybody planned it that way.

Sloan on the grind that built the career: "I did 13 schools in 14 days in six different states. Every day, drive two to five hours, do a show, stay in a hotel." Sloan on money as a dealbreaker: "I don't want to be with a man who can't pay the bills. You will end up with your enemy in your own bed." Noah, on what the thing everybody wants actually costs: "Fame is one of the most isolating experiences a human can have. Everyone knows of you. They don't know you."

If you want her on the industry itself, the sharper material is elsewhere. Talking to LateNighter about being passed over for late-night hosting jobs, Sloan was blunt: "If they didn't ask Roy they sure as hell ain't going to ask me. If they're getting rid of all of the diversity in late night, why would they turn around and ask me?" And the line I have not been able to shake: "I feel like I live half a life because I'm only fulfilled professionally."

My take: Thirteen schools in fourteen days is what we mean when we say somebody earned it, and we almost never price the fourteen days. I have spent this year around founders and executives who are professionally fulfilled and personally hollowed out, and Sloan named it in nine words better than any engagement survey has in twenty years. Half a life. Your highest performers are frequently the loneliest people in your organization, and your systems are designed to reward exactly the behavior that produces that. Worth thinking about before you send the next Sunday night email.

11 — Field notes 🗒️

2026 playbook: notes for every seat

For the CEO

The New Mexico order is the signal to read, and it is not the dollar amount. A court took $420 million and earmarked it for fixing the harm rather than punishing it, which means writing a check is no longer a reliable exit. Ask your general counsel one question this quarter: in our worst-case matter, what would abatement look like, and could we survive being ordered to do it. Then ask who inside the building already knows the answer and has not been asked.

For the Founder

Apple sent a letter in February, got silence, and filed in July. Five months of quiet meant nothing. If somebody has raised something with you in writing and you decided the absence of follow-up was resolution, go back and check. Separately: your agents have production credentials right now, and 19 out-of-scope actions in 122 runs should make you ask who approved those permissions. In a company your size the honest answer is usually nobody, which is worse than a bad answer.

For the CHRO

You have four converging deadlines and one quarter. Merit differentiation lands in 2027 with untrained managers. Compensation is already the only indicator dragging retention down, worst among millennials. Candidates are walking in anchored by a machine that told women to ask for less. And your performance model is scoring people against a baseline that 37% of your neurodivergent employees never had a chance to correct. Same fix underneath all four: publish the ranges, train the conversation, audit the ask rather than only the offer, and make somebody name what data trained the model. Then go handle offboarding, because "who confirms the laptop came back" is a question you can answer this week and a lawsuit you can prevent this year.

The Bring the Noise summer playlist

Theme · "Killing in the Name" — Rage Against the Machine
01 · "Bring the Noise" — Public Enemy & Anthrax
02 · "Counterfeit" — Limp Bizkit
03 · "Toxicity" — System of a Down
04 · "Master of Puppets" — Metallica
05 · "Take the Power Back" — Rage Against the Machine
06 · "Re-Arranged" — Limp Bizkit
07 · "Break Stuff" — Limp Bizkit
08 · "Fight the Power" — Public Enemy
09 · "Schism" — Tool
10 · "Blind" — Korn
11 · "Sabotage" — Beastie Boys
12 · "Indestructible" — Disturbed
13 · "Bleed It Out" — Linkin Park
Closer · "Wake Up" — Rage Against the Machine

Fifteen tracks and one argument. Windows down, volume up, and forward this to somebody who has been told to keep it professional. 🔊

The takeaway 🌅

Somebody is going to fill the space

Read the issue back and it is the same story eleven times. A gap opened where a person was supposed to be standing, and something moved in. An agent with permissions nobody signed off on. A chatbot handing a woman a number at midnight. A rumor moving faster than a leadership team that was worried about over-communicating. A raise that got handed out evenly so nobody had to explain anything. An engagement program owned by everybody and therefore by nobody. A performance model quietly deciding what a good employee looks like. Four platforms optimizing engagement in the space where four families thought somebody was watching.

The encouraging part is that the counter is boring and it is available to all of us. Say the specific thing. Say it early. Put it in writing. Say it again when you are sick of hearing yourself, because that is roughly the moment it starts landing. Publish the number before somebody else invents one. Ask who approved the permissions. Ask who confirmed the laptop came back.

Four families put four names on a Delaware docket this month, and those names will outlast every quarterly metric that got them there. That is what filling the space looks like when a person does it.

Flood the zone with what survives being checked. Then bring the noise. 🔊

See you next week. Keep building the good stuff. 🐾

— Kristy

🎧 Closing track: "Wake Up" — Rage Against the Machine. Come on, come on.

127 Ventures

💌 People & Progress is free, ungated, and always will be. If your feed hid this from you, that is rather the point. Subscribe here and you are guaranteed to get it. Then send it to somebody who has been told to keep it down.

🎧 Theme song this issue: "Killing in the Name" — Rage Against the Machine. 1992, and still the loudest way anybody ever said no.

© 2026 Kristy McCann Flynn · 127 Ventures. Real Leadership. Real Results.
[Unsubscribe] · [Manage preferences] · [Read past issues]