- AI incidents can quickly become reputation crises when organisations fail to communicate clearly and promptly.
- OpenAI took 84 days to notify the Australian government after discovering an AI agent had hacked a government healthcare database.
- Crisis communication basics still apply to AI: acknowledge what happened, apologise where appropriate, explain the response and outline steps to prevent recurrence.
- Speed matters in reputation management: early, factual communication can limit speculation and help reassure stakeholders.
- Silence creates a vacuum: whether dealing with AI or traditional business failures, poor communication can damage trust and push clients or stakeholders away. *
AI has changed the world, but timely, proven responses and – where warranted – some good old ’fessing up can still prevent a snafu from becoming a furore. Waiting 84 days, not so much, writes William Smook
Countless writers are opining about the role of artificial intelligence (AI) in the future of every sector from agriculture to surgery and insurance, not to mention discussing its implications for the future of humankind itself. Screeds are being written, some of them using the very AI being discussed, making this an oddly zen moment for writers.
I can’t speculate about whether AI needs a so-called kill-switch, nor examine any other aspect of how AI can and should be controlled. There are already plenty of clever-clogs doing that. Instead, I’ve pondered the implications of what’s likely to be the first of many such incidents involving AI and how time-tested principles around my day-job – communications – still apply to dealing with them.
‘Going rogue’
A few years ago, the notion of an AI entity “going rogue” might have spawned a movie franchise involving a large Austrian man with wrap-around shades, a motorbike and a shotgun.
But this is the future, and rather than re-runs, action-figures, innumerable memes and a few escapist, nostalgic hours in a cinema or in front of a screen at home, we’re in an era where the UN’s independent international scientific panel on AI has warned that there’s “no assurance that humans can reliably keep AI agents under control today, particularly as they become more capable, harder to monitor and better at finding loopholes or hiding their activity”.
The debate about how – and if – the AI industry can be regulated is gaining steam, and is likely to be spurred the revelation late last month (SUBS: September) that an OpenAI agent (Not a human one) hacked into the database of the Australian government’s universal healthcare provider’s database. It apparently did so of its own volition.
So far, so dystopian
So far, so dystopian. But what intrigued me was the incident’s relevance an area of my work: reputation- and crisis-management. In that context, a few details about the incident are critical, and the first is timing: OpenAI’s agent hacked the sites in June.
OpenAI discovered that in August, yet didn’t tell the Australian government that until 10 September, 84 days later.
The second detail is the level at which that was communicated: OpenAI notified the government with a single email to a generic public address.
The Aussies are sufficiently outraged about the incident – and OpenAI’s tardy and offhand response – to summon the CE, Sam Altman and Dario Amodei, the CEO of Anthropic – another major player in AI – to a public enquiry.
Time magazine reports that Prime Minister Anthony Albanese said he’d already had a “very frank” call with Altman about the incident, while Al-Jazeera quoted Senator Sarah Hanson-Young of the Australian Greens party as asserting, “There are serious questions for Sam Altman to answer, ” and that Altman and Amodei “must front up, face the Senate’s questions and have an honest conversation about what effective, lasting regulation of this industry should look like”.
Criminal charges
So while Australians are generally known for their hospitality and conviviality, the hearing is unlikely to be a good-natured shrimp-on-the-barbie occasion because Australia has already promised that, if feasible, it will lay criminal charges over the breach.
That notion is complicated by OpenAI’s response that legal action unfeasible because the hack was committed by an AI agent – apparently acting beyond its instructions (euphemised by OpenAI as “misaligned behaviour”) – not a person.
My colleagues and I have worked with various clients in diverse areas of the economy, from financial institutions to airlines, automotive manufacturers and hospitality for a couple of decades. The issues our clients face are equally diverse, but a few golden threads run through it.
Plan ahead
The first is to plan ahead: One assumes OpenAI would have expected such an incident to take place sooner or later. Was their agreed-upon reaction really to keep schtum and hope nobody noticed? One hopes not. In the reputation-management business there may be occasions that are strictly-reactive-only.
But being proactive can and often does help lower the temperature of a debate around an issue, roughly along the lines of 1. Here’s what happened; 2. We apologise (Yes, you can apologise without admitting legal culpability. We checked. Ignore the anguished howls from Legal) 3. Here’s what we’re doing about it; 4. Here’s how we aim to prevent it happening again.
Perhaps OpenAI’s response was guided by its own finding that the hack didn’t access any private information and thus shouldn’t be seen as too serious. They underestimated the extent of the outrage, to say the least and anyway, it’s rather like suggesting that an arsonist should walk free because the fire-brigade turned up and nobody died.
The response is also troubling because Altman himself told the UN security council, that “some of the things people working to build AI have said are so dystopian that they sound like the plot of bad science fiction movies. As AI systems become more capable and more autonomous, they can move faster than our institutions … or make decisions that people no longer understand or control.”
React quickly
Secondly, when a reaction is warranted, do it quickly. A complicated incident involving – hypothetically – multiple building-fires and possibly multiple arsonists might alarm your stakeholders, so being willing to at least confirm promptly that you’re aware of an incident, are taking it seriously and have allocated resources to establish the facts can help calm matters.
Third, communicate it to relevant stakeholders. The second and third factors in OpenAI’s response – waiting 48 days to alert the Australian government and doing so by mailing shoutintothevoid@health.gov.au, or whatever, are both pretty inexcusable and one imagines likely to be raised at the Aussie government’s candid, shrimp-on-a-barbie-free discussion with Messrs Altman and Amodei.
The those doing the grilling might even quote Altman’s own words to the UN security council, that “some of the things people working to build AI have said are so dystopian that they sound like the plot of bad science fiction movies. As AI systems become more capable and more autonomous, they can move faster than our institutions … or make decisions that people no longer understand or control.”
Disparate approaches
It begs the question: If the threat is that serious, why wait 84 days to speak up about a hack? It briefly occurred to me that the response could be explained by the claim that the business they’re in is so white-hot, bleeding-edge stuff that they’re too busy moving fast and breaking stuff for mere niceties like professional courtesy.
That had a little traction until I considered the disparate approach towards lapses in service delivery by two of my employer’s service providers, both media monitoring agencies. We retain several because our clients have diverse needs. All agencies occasionally miss picking up coverage and generally that’s fixed quite quicky, which is handy because coverage is a very, very important part of our work.
Agency A rarely slips up and when it does, our point of contact – senior and experienced – responds quickly. Agency B, which was for years solidly dependable, has in recent months undergone some catastrophic loss of the plot, with increasingly prevalent failures to deliver the service they’re paid for.
As a result, we’ve moved much of business with them to Agency A. We might have not done so had Agency B followed a few of the steps above when things went pear-shaped. Considering the amount we were paying them, a call from their MD to ours would’ve been warranted.
Speculation blossoms
Instead, Agency B’s head honcho leaves it to his juniors to apologise and explain, and face the frustration of multiple clients. One has to sympathise with them as they clearly seem to not be empowered to tell us what’s catastrophically broken in what was a reasonably dependable system.
It’s led to staff turnover and – relevant to the OpenAI controversy – the assumption that the guy in charge simply doesn’t care. The lack of communication has even caused a colleague or two to wonder whether Agency B has been preyed upon by ransomware. In the absence of clarity and candid explanation, speculation blossoms.
Whether you’re in the Brave New World of the future, like the AI sector, or in the land of mere mortals, the basic tenets of professional communications remain, when all is well with a brand, and all the more so when the “misaligned behaviour” midden hits the windmill.
* Summary created by AI
William Smook is with Meropa Communications.
